1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * linux/fs/jbd2/journal.c 4 * 5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 6 * 7 * Copyright 1998 Red Hat corp --- All Rights Reserved 8 * 9 * Generic filesystem journal-writing code; part of the ext2fs 10 * journaling system. 11 * 12 * This file manages journals: areas of disk reserved for logging 13 * transactional updates. This includes the kernel journaling thread 14 * which is responsible for scheduling updates to the log. 15 * 16 * We do not actually manage the physical storage of the journal in this 17 * file: that is left to a per-journal policy function, which allows us 18 * to store the journal within a filesystem-specified area for ext2 19 * journaling (ext2 can use a reserved inode for storing the log). 20 */ 21 22 #include <linux/module.h> 23 #include <linux/time.h> 24 #include <linux/fs.h> 25 #include <linux/jbd2.h> 26 #include <linux/errno.h> 27 #include <linux/slab.h> 28 #include <linux/init.h> 29 #include <linux/mm.h> 30 #include <linux/freezer.h> 31 #include <linux/pagemap.h> 32 #include <linux/kthread.h> 33 #include <linux/poison.h> 34 #include <linux/proc_fs.h> 35 #include <linux/seq_file.h> 36 #include <linux/math64.h> 37 #include <linux/hash.h> 38 #include <linux/log2.h> 39 #include <linux/vmalloc.h> 40 #include <linux/backing-dev.h> 41 #include <linux/bitops.h> 42 #include <linux/ratelimit.h> 43 #include <linux/sched/mm.h> 44 45 #define CREATE_TRACE_POINTS 46 #include <trace/events/jbd2.h> 47 48 #include <linux/uaccess.h> 49 #include <asm/page.h> 50 51 #ifdef CONFIG_JBD2_DEBUG 52 static ushort jbd2_journal_enable_debug __read_mostly; 53 54 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644); 55 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2"); 56 #endif 57 58 EXPORT_SYMBOL(jbd2_journal_extend); 59 EXPORT_SYMBOL(jbd2_journal_stop); 60 EXPORT_SYMBOL(jbd2_journal_lock_updates); 61 EXPORT_SYMBOL(jbd2_journal_unlock_updates); 62 EXPORT_SYMBOL(jbd2_journal_get_write_access); 63 EXPORT_SYMBOL(jbd2_journal_get_create_access); 64 EXPORT_SYMBOL(jbd2_journal_get_undo_access); 65 EXPORT_SYMBOL(jbd2_journal_set_triggers); 66 EXPORT_SYMBOL(jbd2_journal_dirty_metadata); 67 EXPORT_SYMBOL(jbd2_journal_forget); 68 EXPORT_SYMBOL(jbd2_journal_flush); 69 EXPORT_SYMBOL(jbd2_journal_revoke); 70 71 EXPORT_SYMBOL(jbd2_journal_init_dev); 72 EXPORT_SYMBOL(jbd2_journal_init_inode); 73 EXPORT_SYMBOL(jbd2_journal_check_used_features); 74 EXPORT_SYMBOL(jbd2_journal_check_available_features); 75 EXPORT_SYMBOL(jbd2_journal_set_features); 76 EXPORT_SYMBOL(jbd2_journal_load); 77 EXPORT_SYMBOL(jbd2_journal_destroy); 78 EXPORT_SYMBOL(jbd2_journal_abort); 79 EXPORT_SYMBOL(jbd2_journal_errno); 80 EXPORT_SYMBOL(jbd2_journal_ack_err); 81 EXPORT_SYMBOL(jbd2_journal_clear_err); 82 EXPORT_SYMBOL(jbd2_log_wait_commit); 83 EXPORT_SYMBOL(jbd2_journal_start_commit); 84 EXPORT_SYMBOL(jbd2_journal_force_commit_nested); 85 EXPORT_SYMBOL(jbd2_journal_wipe); 86 EXPORT_SYMBOL(jbd2_journal_blocks_per_page); 87 EXPORT_SYMBOL(jbd2_journal_invalidate_folio); 88 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); 89 EXPORT_SYMBOL(jbd2_journal_force_commit); 90 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write); 91 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait); 92 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers); 93 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); 94 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); 95 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); 96 EXPORT_SYMBOL(jbd2_inode_cache); 97 98 static int jbd2_journal_create_slab(size_t slab_size); 99 100 #ifdef CONFIG_JBD2_DEBUG 101 void __jbd2_debug(int level, const char *file, const char *func, 102 unsigned int line, const char *fmt, ...) 103 { 104 struct va_format vaf; 105 va_list args; 106 107 if (level > jbd2_journal_enable_debug) 108 return; 109 va_start(args, fmt); 110 vaf.fmt = fmt; 111 vaf.va = &args; 112 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf); 113 va_end(args); 114 } 115 #endif 116 117 /* Checksumming functions */ 118 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) 119 { 120 if (!jbd2_journal_has_csum_v2or3_feature(j)) 121 return 1; 122 123 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; 124 } 125 126 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) 127 { 128 __u32 csum; 129 __be32 old_csum; 130 131 old_csum = sb->s_checksum; 132 sb->s_checksum = 0; 133 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); 134 sb->s_checksum = old_csum; 135 136 return cpu_to_be32(csum); 137 } 138 139 /* 140 * Helper function used to manage commit timeouts 141 */ 142 143 static void commit_timeout(struct timer_list *t) 144 { 145 journal_t *journal = from_timer(journal, t, j_commit_timer); 146 147 wake_up_process(journal->j_task); 148 } 149 150 /* 151 * kjournald2: The main thread function used to manage a logging device 152 * journal. 153 * 154 * This kernel thread is responsible for two things: 155 * 156 * 1) COMMIT: Every so often we need to commit the current state of the 157 * filesystem to disk. The journal thread is responsible for writing 158 * all of the metadata buffers to disk. If a fast commit is ongoing 159 * journal thread waits until it's done and then continues from 160 * there on. 161 * 162 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all 163 * of the data in that part of the log has been rewritten elsewhere on 164 * the disk. Flushing these old buffers to reclaim space in the log is 165 * known as checkpointing, and this thread is responsible for that job. 166 */ 167 168 static int kjournald2(void *arg) 169 { 170 journal_t *journal = arg; 171 transaction_t *transaction; 172 173 /* 174 * Set up an interval timer which can be used to trigger a commit wakeup 175 * after the commit interval expires 176 */ 177 timer_setup(&journal->j_commit_timer, commit_timeout, 0); 178 179 set_freezable(); 180 181 /* Record that the journal thread is running */ 182 journal->j_task = current; 183 wake_up(&journal->j_wait_done_commit); 184 185 /* 186 * Make sure that no allocations from this kernel thread will ever 187 * recurse to the fs layer because we are responsible for the 188 * transaction commit and any fs involvement might get stuck waiting for 189 * the trasn. commit. 190 */ 191 memalloc_nofs_save(); 192 193 /* 194 * And now, wait forever for commit wakeup events. 195 */ 196 write_lock(&journal->j_state_lock); 197 198 loop: 199 if (journal->j_flags & JBD2_UNMOUNT) 200 goto end_loop; 201 202 jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n", 203 journal->j_commit_sequence, journal->j_commit_request); 204 205 if (journal->j_commit_sequence != journal->j_commit_request) { 206 jbd2_debug(1, "OK, requests differ\n"); 207 write_unlock(&journal->j_state_lock); 208 del_timer_sync(&journal->j_commit_timer); 209 jbd2_journal_commit_transaction(journal); 210 write_lock(&journal->j_state_lock); 211 goto loop; 212 } 213 214 wake_up(&journal->j_wait_done_commit); 215 if (freezing(current)) { 216 /* 217 * The simpler the better. Flushing journal isn't a 218 * good idea, because that depends on threads that may 219 * be already stopped. 220 */ 221 jbd2_debug(1, "Now suspending kjournald2\n"); 222 write_unlock(&journal->j_state_lock); 223 try_to_freeze(); 224 write_lock(&journal->j_state_lock); 225 } else { 226 /* 227 * We assume on resume that commits are already there, 228 * so we don't sleep 229 */ 230 DEFINE_WAIT(wait); 231 int should_sleep = 1; 232 233 prepare_to_wait(&journal->j_wait_commit, &wait, 234 TASK_INTERRUPTIBLE); 235 if (journal->j_commit_sequence != journal->j_commit_request) 236 should_sleep = 0; 237 transaction = journal->j_running_transaction; 238 if (transaction && time_after_eq(jiffies, 239 transaction->t_expires)) 240 should_sleep = 0; 241 if (journal->j_flags & JBD2_UNMOUNT) 242 should_sleep = 0; 243 if (should_sleep) { 244 write_unlock(&journal->j_state_lock); 245 schedule(); 246 write_lock(&journal->j_state_lock); 247 } 248 finish_wait(&journal->j_wait_commit, &wait); 249 } 250 251 jbd2_debug(1, "kjournald2 wakes\n"); 252 253 /* 254 * Were we woken up by a commit wakeup event? 255 */ 256 transaction = journal->j_running_transaction; 257 if (transaction && time_after_eq(jiffies, transaction->t_expires)) { 258 journal->j_commit_request = transaction->t_tid; 259 jbd2_debug(1, "woke because of timeout\n"); 260 } 261 goto loop; 262 263 end_loop: 264 del_timer_sync(&journal->j_commit_timer); 265 journal->j_task = NULL; 266 wake_up(&journal->j_wait_done_commit); 267 jbd2_debug(1, "Journal thread exiting.\n"); 268 write_unlock(&journal->j_state_lock); 269 return 0; 270 } 271 272 static int jbd2_journal_start_thread(journal_t *journal) 273 { 274 struct task_struct *t; 275 276 t = kthread_run(kjournald2, journal, "jbd2/%s", 277 journal->j_devname); 278 if (IS_ERR(t)) 279 return PTR_ERR(t); 280 281 wait_event(journal->j_wait_done_commit, journal->j_task != NULL); 282 return 0; 283 } 284 285 static void journal_kill_thread(journal_t *journal) 286 { 287 write_lock(&journal->j_state_lock); 288 journal->j_flags |= JBD2_UNMOUNT; 289 290 while (journal->j_task) { 291 write_unlock(&journal->j_state_lock); 292 wake_up(&journal->j_wait_commit); 293 wait_event(journal->j_wait_done_commit, journal->j_task == NULL); 294 write_lock(&journal->j_state_lock); 295 } 296 write_unlock(&journal->j_state_lock); 297 } 298 299 /* 300 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. 301 * 302 * Writes a metadata buffer to a given disk block. The actual IO is not 303 * performed but a new buffer_head is constructed which labels the data 304 * to be written with the correct destination disk block. 305 * 306 * Any magic-number escaping which needs to be done will cause a 307 * copy-out here. If the buffer happens to start with the 308 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the 309 * magic number is only written to the log for descripter blocks. In 310 * this case, we copy the data and replace the first word with 0, and we 311 * return a result code which indicates that this buffer needs to be 312 * marked as an escaped buffer in the corresponding log descriptor 313 * block. The missing word can then be restored when the block is read 314 * during recovery. 315 * 316 * If the source buffer has already been modified by a new transaction 317 * since we took the last commit snapshot, we use the frozen copy of 318 * that data for IO. If we end up using the existing buffer_head's data 319 * for the write, then we have to make sure nobody modifies it while the 320 * IO is in progress. do_get_write_access() handles this. 321 * 322 * The function returns a pointer to the buffer_head to be used for IO. 323 * 324 * 325 * Return value: 326 * <0: Error 327 * >=0: Finished OK 328 * 329 * On success: 330 * Bit 0 set == escape performed on the data 331 * Bit 1 set == buffer copy-out performed (kfree the data after IO) 332 */ 333 334 int jbd2_journal_write_metadata_buffer(transaction_t *transaction, 335 struct journal_head *jh_in, 336 struct buffer_head **bh_out, 337 sector_t blocknr) 338 { 339 int need_copy_out = 0; 340 int done_copy_out = 0; 341 int do_escape = 0; 342 char *mapped_data; 343 struct buffer_head *new_bh; 344 struct folio *new_folio; 345 unsigned int new_offset; 346 struct buffer_head *bh_in = jh2bh(jh_in); 347 journal_t *journal = transaction->t_journal; 348 349 /* 350 * The buffer really shouldn't be locked: only the current committing 351 * transaction is allowed to write it, so nobody else is allowed 352 * to do any IO. 353 * 354 * akpm: except if we're journalling data, and write() output is 355 * also part of a shared mapping, and another thread has 356 * decided to launch a writepage() against this buffer. 357 */ 358 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); 359 360 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); 361 362 /* keep subsequent assertions sane */ 363 atomic_set(&new_bh->b_count, 1); 364 365 spin_lock(&jh_in->b_state_lock); 366 repeat: 367 /* 368 * If a new transaction has already done a buffer copy-out, then 369 * we use that version of the data for the commit. 370 */ 371 if (jh_in->b_frozen_data) { 372 done_copy_out = 1; 373 new_folio = virt_to_folio(jh_in->b_frozen_data); 374 new_offset = offset_in_folio(new_folio, jh_in->b_frozen_data); 375 } else { 376 new_folio = jh2bh(jh_in)->b_folio; 377 new_offset = offset_in_folio(new_folio, jh2bh(jh_in)->b_data); 378 } 379 380 mapped_data = kmap_local_folio(new_folio, new_offset); 381 /* 382 * Fire data frozen trigger if data already wasn't frozen. Do this 383 * before checking for escaping, as the trigger may modify the magic 384 * offset. If a copy-out happens afterwards, it will have the correct 385 * data in the buffer. 386 */ 387 if (!done_copy_out) 388 jbd2_buffer_frozen_trigger(jh_in, mapped_data, 389 jh_in->b_triggers); 390 391 /* 392 * Check for escaping 393 */ 394 if (*((__be32 *)mapped_data) == cpu_to_be32(JBD2_MAGIC_NUMBER)) { 395 need_copy_out = 1; 396 do_escape = 1; 397 } 398 kunmap_local(mapped_data); 399 400 /* 401 * Do we need to do a data copy? 402 */ 403 if (need_copy_out && !done_copy_out) { 404 char *tmp; 405 406 spin_unlock(&jh_in->b_state_lock); 407 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); 408 if (!tmp) { 409 brelse(new_bh); 410 return -ENOMEM; 411 } 412 spin_lock(&jh_in->b_state_lock); 413 if (jh_in->b_frozen_data) { 414 jbd2_free(tmp, bh_in->b_size); 415 goto repeat; 416 } 417 418 jh_in->b_frozen_data = tmp; 419 memcpy_from_folio(tmp, new_folio, new_offset, bh_in->b_size); 420 421 new_folio = virt_to_folio(tmp); 422 new_offset = offset_in_folio(new_folio, tmp); 423 done_copy_out = 1; 424 425 /* 426 * This isn't strictly necessary, as we're using frozen 427 * data for the escaping, but it keeps consistency with 428 * b_frozen_data usage. 429 */ 430 jh_in->b_frozen_triggers = jh_in->b_triggers; 431 } 432 433 /* 434 * Did we need to do an escaping? Now we've done all the 435 * copying, we can finally do so. 436 */ 437 if (do_escape) { 438 mapped_data = kmap_local_folio(new_folio, new_offset); 439 *((unsigned int *)mapped_data) = 0; 440 kunmap_local(mapped_data); 441 } 442 443 folio_set_bh(new_bh, new_folio, new_offset); 444 new_bh->b_size = bh_in->b_size; 445 new_bh->b_bdev = journal->j_dev; 446 new_bh->b_blocknr = blocknr; 447 new_bh->b_private = bh_in; 448 set_buffer_mapped(new_bh); 449 set_buffer_dirty(new_bh); 450 451 *bh_out = new_bh; 452 453 /* 454 * The to-be-written buffer needs to get moved to the io queue, 455 * and the original buffer whose contents we are shadowing or 456 * copying is moved to the transaction's shadow queue. 457 */ 458 JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); 459 spin_lock(&journal->j_list_lock); 460 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); 461 spin_unlock(&journal->j_list_lock); 462 set_buffer_shadow(bh_in); 463 spin_unlock(&jh_in->b_state_lock); 464 465 return do_escape | (done_copy_out << 1); 466 } 467 468 /* 469 * Allocation code for the journal file. Manage the space left in the 470 * journal, so that we can begin checkpointing when appropriate. 471 */ 472 473 /* 474 * Called with j_state_lock locked for writing. 475 * Returns true if a transaction commit was started. 476 */ 477 static int __jbd2_log_start_commit(journal_t *journal, tid_t target) 478 { 479 /* Return if the txn has already requested to be committed */ 480 if (journal->j_commit_request == target) 481 return 0; 482 483 /* 484 * The only transaction we can possibly wait upon is the 485 * currently running transaction (if it exists). Otherwise, 486 * the target tid must be an old one. 487 */ 488 if (journal->j_running_transaction && 489 journal->j_running_transaction->t_tid == target) { 490 /* 491 * We want a new commit: OK, mark the request and wakeup the 492 * commit thread. We do _not_ do the commit ourselves. 493 */ 494 495 journal->j_commit_request = target; 496 jbd2_debug(1, "JBD2: requesting commit %u/%u\n", 497 journal->j_commit_request, 498 journal->j_commit_sequence); 499 journal->j_running_transaction->t_requested = jiffies; 500 wake_up(&journal->j_wait_commit); 501 return 1; 502 } else if (!tid_geq(journal->j_commit_request, target)) 503 /* This should never happen, but if it does, preserve 504 the evidence before kjournald goes into a loop and 505 increments j_commit_sequence beyond all recognition. */ 506 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", 507 journal->j_commit_request, 508 journal->j_commit_sequence, 509 target, journal->j_running_transaction ? 510 journal->j_running_transaction->t_tid : 0); 511 return 0; 512 } 513 514 int jbd2_log_start_commit(journal_t *journal, tid_t tid) 515 { 516 int ret; 517 518 write_lock(&journal->j_state_lock); 519 ret = __jbd2_log_start_commit(journal, tid); 520 write_unlock(&journal->j_state_lock); 521 return ret; 522 } 523 524 /* 525 * Force and wait any uncommitted transactions. We can only force the running 526 * transaction if we don't have an active handle, otherwise, we will deadlock. 527 * Returns: <0 in case of error, 528 * 0 if nothing to commit, 529 * 1 if transaction was successfully committed. 530 */ 531 static int __jbd2_journal_force_commit(journal_t *journal) 532 { 533 transaction_t *transaction = NULL; 534 tid_t tid; 535 int need_to_start = 0, ret = 0; 536 537 read_lock(&journal->j_state_lock); 538 if (journal->j_running_transaction && !current->journal_info) { 539 transaction = journal->j_running_transaction; 540 if (!tid_geq(journal->j_commit_request, transaction->t_tid)) 541 need_to_start = 1; 542 } else if (journal->j_committing_transaction) 543 transaction = journal->j_committing_transaction; 544 545 if (!transaction) { 546 /* Nothing to commit */ 547 read_unlock(&journal->j_state_lock); 548 return 0; 549 } 550 tid = transaction->t_tid; 551 read_unlock(&journal->j_state_lock); 552 if (need_to_start) 553 jbd2_log_start_commit(journal, tid); 554 ret = jbd2_log_wait_commit(journal, tid); 555 if (!ret) 556 ret = 1; 557 558 return ret; 559 } 560 561 /** 562 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the 563 * calling process is not within transaction. 564 * 565 * @journal: journal to force 566 * Returns true if progress was made. 567 * 568 * This is used for forcing out undo-protected data which contains 569 * bitmaps, when the fs is running out of space. 570 */ 571 int jbd2_journal_force_commit_nested(journal_t *journal) 572 { 573 int ret; 574 575 ret = __jbd2_journal_force_commit(journal); 576 return ret > 0; 577 } 578 579 /** 580 * jbd2_journal_force_commit() - force any uncommitted transactions 581 * @journal: journal to force 582 * 583 * Caller want unconditional commit. We can only force the running transaction 584 * if we don't have an active handle, otherwise, we will deadlock. 585 */ 586 int jbd2_journal_force_commit(journal_t *journal) 587 { 588 int ret; 589 590 J_ASSERT(!current->journal_info); 591 ret = __jbd2_journal_force_commit(journal); 592 if (ret > 0) 593 ret = 0; 594 return ret; 595 } 596 597 /* 598 * Start a commit of the current running transaction (if any). Returns true 599 * if a transaction is going to be committed (or is currently already 600 * committing), and fills its tid in at *ptid 601 */ 602 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) 603 { 604 int ret = 0; 605 606 write_lock(&journal->j_state_lock); 607 if (journal->j_running_transaction) { 608 tid_t tid = journal->j_running_transaction->t_tid; 609 610 __jbd2_log_start_commit(journal, tid); 611 /* There's a running transaction and we've just made sure 612 * it's commit has been scheduled. */ 613 if (ptid) 614 *ptid = tid; 615 ret = 1; 616 } else if (journal->j_committing_transaction) { 617 /* 618 * If commit has been started, then we have to wait for 619 * completion of that transaction. 620 */ 621 if (ptid) 622 *ptid = journal->j_committing_transaction->t_tid; 623 ret = 1; 624 } 625 write_unlock(&journal->j_state_lock); 626 return ret; 627 } 628 629 /* 630 * Return 1 if a given transaction has not yet sent barrier request 631 * connected with a transaction commit. If 0 is returned, transaction 632 * may or may not have sent the barrier. Used to avoid sending barrier 633 * twice in common cases. 634 */ 635 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) 636 { 637 int ret = 0; 638 transaction_t *commit_trans; 639 640 if (!(journal->j_flags & JBD2_BARRIER)) 641 return 0; 642 read_lock(&journal->j_state_lock); 643 /* Transaction already committed? */ 644 if (tid_geq(journal->j_commit_sequence, tid)) 645 goto out; 646 commit_trans = journal->j_committing_transaction; 647 if (!commit_trans || commit_trans->t_tid != tid) { 648 ret = 1; 649 goto out; 650 } 651 /* 652 * Transaction is being committed and we already proceeded to 653 * submitting a flush to fs partition? 654 */ 655 if (journal->j_fs_dev != journal->j_dev) { 656 if (!commit_trans->t_need_data_flush || 657 commit_trans->t_state >= T_COMMIT_DFLUSH) 658 goto out; 659 } else { 660 if (commit_trans->t_state >= T_COMMIT_JFLUSH) 661 goto out; 662 } 663 ret = 1; 664 out: 665 read_unlock(&journal->j_state_lock); 666 return ret; 667 } 668 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); 669 670 /* 671 * Wait for a specified commit to complete. 672 * The caller may not hold the journal lock. 673 */ 674 int jbd2_log_wait_commit(journal_t *journal, tid_t tid) 675 { 676 int err = 0; 677 678 read_lock(&journal->j_state_lock); 679 #ifdef CONFIG_PROVE_LOCKING 680 /* 681 * Some callers make sure transaction is already committing and in that 682 * case we cannot block on open handles anymore. So don't warn in that 683 * case. 684 */ 685 if (tid_gt(tid, journal->j_commit_sequence) && 686 (!journal->j_committing_transaction || 687 journal->j_committing_transaction->t_tid != tid)) { 688 read_unlock(&journal->j_state_lock); 689 jbd2_might_wait_for_commit(journal); 690 read_lock(&journal->j_state_lock); 691 } 692 #endif 693 #ifdef CONFIG_JBD2_DEBUG 694 if (!tid_geq(journal->j_commit_request, tid)) { 695 printk(KERN_ERR 696 "%s: error: j_commit_request=%u, tid=%u\n", 697 __func__, journal->j_commit_request, tid); 698 } 699 #endif 700 while (tid_gt(tid, journal->j_commit_sequence)) { 701 jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n", 702 tid, journal->j_commit_sequence); 703 read_unlock(&journal->j_state_lock); 704 wake_up(&journal->j_wait_commit); 705 wait_event(journal->j_wait_done_commit, 706 !tid_gt(tid, journal->j_commit_sequence)); 707 read_lock(&journal->j_state_lock); 708 } 709 read_unlock(&journal->j_state_lock); 710 711 if (unlikely(is_journal_aborted(journal))) 712 err = -EIO; 713 return err; 714 } 715 716 /* 717 * Start a fast commit. If there's an ongoing fast or full commit wait for 718 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY 719 * if a fast commit is not needed, either because there's an already a commit 720 * going on or this tid has already been committed. Returns -EINVAL if no jbd2 721 * commit has yet been performed. 722 */ 723 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid) 724 { 725 if (unlikely(is_journal_aborted(journal))) 726 return -EIO; 727 /* 728 * Fast commits only allowed if at least one full commit has 729 * been processed. 730 */ 731 if (!journal->j_stats.ts_tid) 732 return -EINVAL; 733 734 write_lock(&journal->j_state_lock); 735 if (tid <= journal->j_commit_sequence) { 736 write_unlock(&journal->j_state_lock); 737 return -EALREADY; 738 } 739 740 if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING || 741 (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) { 742 DEFINE_WAIT(wait); 743 744 prepare_to_wait(&journal->j_fc_wait, &wait, 745 TASK_UNINTERRUPTIBLE); 746 write_unlock(&journal->j_state_lock); 747 schedule(); 748 finish_wait(&journal->j_fc_wait, &wait); 749 return -EALREADY; 750 } 751 journal->j_flags |= JBD2_FAST_COMMIT_ONGOING; 752 write_unlock(&journal->j_state_lock); 753 jbd2_journal_lock_updates(journal); 754 755 return 0; 756 } 757 EXPORT_SYMBOL(jbd2_fc_begin_commit); 758 759 /* 760 * Stop a fast commit. If fallback is set, this function starts commit of 761 * TID tid before any other fast commit can start. 762 */ 763 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback) 764 { 765 jbd2_journal_unlock_updates(journal); 766 if (journal->j_fc_cleanup_callback) 767 journal->j_fc_cleanup_callback(journal, 0, tid); 768 write_lock(&journal->j_state_lock); 769 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING; 770 if (fallback) 771 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING; 772 write_unlock(&journal->j_state_lock); 773 wake_up(&journal->j_fc_wait); 774 if (fallback) 775 return jbd2_complete_transaction(journal, tid); 776 return 0; 777 } 778 779 int jbd2_fc_end_commit(journal_t *journal) 780 { 781 return __jbd2_fc_end_commit(journal, 0, false); 782 } 783 EXPORT_SYMBOL(jbd2_fc_end_commit); 784 785 int jbd2_fc_end_commit_fallback(journal_t *journal) 786 { 787 tid_t tid; 788 789 read_lock(&journal->j_state_lock); 790 tid = journal->j_running_transaction ? 791 journal->j_running_transaction->t_tid : 0; 792 read_unlock(&journal->j_state_lock); 793 return __jbd2_fc_end_commit(journal, tid, true); 794 } 795 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback); 796 797 /* Return 1 when transaction with given tid has already committed. */ 798 int jbd2_transaction_committed(journal_t *journal, tid_t tid) 799 { 800 int ret = 1; 801 802 read_lock(&journal->j_state_lock); 803 if (journal->j_running_transaction && 804 journal->j_running_transaction->t_tid == tid) 805 ret = 0; 806 if (journal->j_committing_transaction && 807 journal->j_committing_transaction->t_tid == tid) 808 ret = 0; 809 read_unlock(&journal->j_state_lock); 810 return ret; 811 } 812 EXPORT_SYMBOL(jbd2_transaction_committed); 813 814 /* 815 * When this function returns the transaction corresponding to tid 816 * will be completed. If the transaction has currently running, start 817 * committing that transaction before waiting for it to complete. If 818 * the transaction id is stale, it is by definition already completed, 819 * so just return SUCCESS. 820 */ 821 int jbd2_complete_transaction(journal_t *journal, tid_t tid) 822 { 823 int need_to_wait = 1; 824 825 read_lock(&journal->j_state_lock); 826 if (journal->j_running_transaction && 827 journal->j_running_transaction->t_tid == tid) { 828 if (journal->j_commit_request != tid) { 829 /* transaction not yet started, so request it */ 830 read_unlock(&journal->j_state_lock); 831 jbd2_log_start_commit(journal, tid); 832 goto wait_commit; 833 } 834 } else if (!(journal->j_committing_transaction && 835 journal->j_committing_transaction->t_tid == tid)) 836 need_to_wait = 0; 837 read_unlock(&journal->j_state_lock); 838 if (!need_to_wait) 839 return 0; 840 wait_commit: 841 return jbd2_log_wait_commit(journal, tid); 842 } 843 EXPORT_SYMBOL(jbd2_complete_transaction); 844 845 /* 846 * Log buffer allocation routines: 847 */ 848 849 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) 850 { 851 unsigned long blocknr; 852 853 write_lock(&journal->j_state_lock); 854 J_ASSERT(journal->j_free > 1); 855 856 blocknr = journal->j_head; 857 journal->j_head++; 858 journal->j_free--; 859 if (journal->j_head == journal->j_last) 860 journal->j_head = journal->j_first; 861 write_unlock(&journal->j_state_lock); 862 return jbd2_journal_bmap(journal, blocknr, retp); 863 } 864 865 /* Map one fast commit buffer for use by the file system */ 866 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out) 867 { 868 unsigned long long pblock; 869 unsigned long blocknr; 870 int ret = 0; 871 struct buffer_head *bh; 872 int fc_off; 873 874 *bh_out = NULL; 875 876 if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) { 877 fc_off = journal->j_fc_off; 878 blocknr = journal->j_fc_first + fc_off; 879 journal->j_fc_off++; 880 } else { 881 ret = -EINVAL; 882 } 883 884 if (ret) 885 return ret; 886 887 ret = jbd2_journal_bmap(journal, blocknr, &pblock); 888 if (ret) 889 return ret; 890 891 bh = __getblk(journal->j_dev, pblock, journal->j_blocksize); 892 if (!bh) 893 return -ENOMEM; 894 895 896 journal->j_fc_wbuf[fc_off] = bh; 897 898 *bh_out = bh; 899 900 return 0; 901 } 902 EXPORT_SYMBOL(jbd2_fc_get_buf); 903 904 /* 905 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf 906 * for completion. 907 */ 908 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks) 909 { 910 struct buffer_head *bh; 911 int i, j_fc_off; 912 913 j_fc_off = journal->j_fc_off; 914 915 /* 916 * Wait in reverse order to minimize chances of us being woken up before 917 * all IOs have completed 918 */ 919 for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) { 920 bh = journal->j_fc_wbuf[i]; 921 wait_on_buffer(bh); 922 /* 923 * Update j_fc_off so jbd2_fc_release_bufs can release remain 924 * buffer head. 925 */ 926 if (unlikely(!buffer_uptodate(bh))) { 927 journal->j_fc_off = i + 1; 928 return -EIO; 929 } 930 put_bh(bh); 931 journal->j_fc_wbuf[i] = NULL; 932 } 933 934 return 0; 935 } 936 EXPORT_SYMBOL(jbd2_fc_wait_bufs); 937 938 int jbd2_fc_release_bufs(journal_t *journal) 939 { 940 struct buffer_head *bh; 941 int i, j_fc_off; 942 943 j_fc_off = journal->j_fc_off; 944 945 for (i = j_fc_off - 1; i >= 0; i--) { 946 bh = journal->j_fc_wbuf[i]; 947 if (!bh) 948 break; 949 put_bh(bh); 950 journal->j_fc_wbuf[i] = NULL; 951 } 952 953 return 0; 954 } 955 EXPORT_SYMBOL(jbd2_fc_release_bufs); 956 957 /* 958 * Conversion of logical to physical block numbers for the journal 959 * 960 * On external journals the journal blocks are identity-mapped, so 961 * this is a no-op. If needed, we can use j_blk_offset - everything is 962 * ready. 963 */ 964 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, 965 unsigned long long *retp) 966 { 967 int err = 0; 968 unsigned long long ret; 969 sector_t block = blocknr; 970 971 if (journal->j_bmap) { 972 err = journal->j_bmap(journal, &block); 973 if (err == 0) 974 *retp = block; 975 } else if (journal->j_inode) { 976 ret = bmap(journal->j_inode, &block); 977 978 if (ret || !block) { 979 printk(KERN_ALERT "%s: journal block not found " 980 "at offset %lu on %s\n", 981 __func__, blocknr, journal->j_devname); 982 err = -EIO; 983 jbd2_journal_abort(journal, err); 984 } else { 985 *retp = block; 986 } 987 988 } else { 989 *retp = blocknr; /* +journal->j_blk_offset */ 990 } 991 return err; 992 } 993 994 /* 995 * We play buffer_head aliasing tricks to write data/metadata blocks to 996 * the journal without copying their contents, but for journal 997 * descriptor blocks we do need to generate bona fide buffers. 998 * 999 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying 1000 * the buffer's contents they really should run flush_dcache_page(bh->b_page). 1001 * But we don't bother doing that, so there will be coherency problems with 1002 * mmaps of blockdevs which hold live JBD-controlled filesystems. 1003 */ 1004 struct buffer_head * 1005 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type) 1006 { 1007 journal_t *journal = transaction->t_journal; 1008 struct buffer_head *bh; 1009 unsigned long long blocknr; 1010 journal_header_t *header; 1011 int err; 1012 1013 err = jbd2_journal_next_log_block(journal, &blocknr); 1014 1015 if (err) 1016 return NULL; 1017 1018 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); 1019 if (!bh) 1020 return NULL; 1021 atomic_dec(&transaction->t_outstanding_credits); 1022 lock_buffer(bh); 1023 memset(bh->b_data, 0, journal->j_blocksize); 1024 header = (journal_header_t *)bh->b_data; 1025 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); 1026 header->h_blocktype = cpu_to_be32(type); 1027 header->h_sequence = cpu_to_be32(transaction->t_tid); 1028 set_buffer_uptodate(bh); 1029 unlock_buffer(bh); 1030 BUFFER_TRACE(bh, "return this buffer"); 1031 return bh; 1032 } 1033 1034 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh) 1035 { 1036 struct jbd2_journal_block_tail *tail; 1037 __u32 csum; 1038 1039 if (!jbd2_journal_has_csum_v2or3(j)) 1040 return; 1041 1042 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - 1043 sizeof(struct jbd2_journal_block_tail)); 1044 tail->t_checksum = 0; 1045 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); 1046 tail->t_checksum = cpu_to_be32(csum); 1047 } 1048 1049 /* 1050 * Return tid of the oldest transaction in the journal and block in the journal 1051 * where the transaction starts. 1052 * 1053 * If the journal is now empty, return which will be the next transaction ID 1054 * we will write and where will that transaction start. 1055 * 1056 * The return value is 0 if journal tail cannot be pushed any further, 1 if 1057 * it can. 1058 */ 1059 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, 1060 unsigned long *block) 1061 { 1062 transaction_t *transaction; 1063 int ret; 1064 1065 read_lock(&journal->j_state_lock); 1066 spin_lock(&journal->j_list_lock); 1067 transaction = journal->j_checkpoint_transactions; 1068 if (transaction) { 1069 *tid = transaction->t_tid; 1070 *block = transaction->t_log_start; 1071 } else if ((transaction = journal->j_committing_transaction) != NULL) { 1072 *tid = transaction->t_tid; 1073 *block = transaction->t_log_start; 1074 } else if ((transaction = journal->j_running_transaction) != NULL) { 1075 *tid = transaction->t_tid; 1076 *block = journal->j_head; 1077 } else { 1078 *tid = journal->j_transaction_sequence; 1079 *block = journal->j_head; 1080 } 1081 ret = tid_gt(*tid, journal->j_tail_sequence); 1082 spin_unlock(&journal->j_list_lock); 1083 read_unlock(&journal->j_state_lock); 1084 1085 return ret; 1086 } 1087 1088 /* 1089 * Update information in journal structure and in on disk journal superblock 1090 * about log tail. This function does not check whether information passed in 1091 * really pushes log tail further. It's responsibility of the caller to make 1092 * sure provided log tail information is valid (e.g. by holding 1093 * j_checkpoint_mutex all the time between computing log tail and calling this 1094 * function as is the case with jbd2_cleanup_journal_tail()). 1095 * 1096 * Requires j_checkpoint_mutex 1097 */ 1098 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1099 { 1100 unsigned long freed; 1101 int ret; 1102 1103 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1104 1105 /* 1106 * We cannot afford for write to remain in drive's caches since as 1107 * soon as we update j_tail, next transaction can start reusing journal 1108 * space and if we lose sb update during power failure we'd replay 1109 * old transaction with possibly newly overwritten data. 1110 */ 1111 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, 1112 REQ_SYNC | REQ_FUA); 1113 if (ret) 1114 goto out; 1115 1116 write_lock(&journal->j_state_lock); 1117 freed = block - journal->j_tail; 1118 if (block < journal->j_tail) 1119 freed += journal->j_last - journal->j_first; 1120 1121 trace_jbd2_update_log_tail(journal, tid, block, freed); 1122 jbd2_debug(1, 1123 "Cleaning journal tail from %u to %u (offset %lu), " 1124 "freeing %lu\n", 1125 journal->j_tail_sequence, tid, block, freed); 1126 1127 journal->j_free += freed; 1128 journal->j_tail_sequence = tid; 1129 journal->j_tail = block; 1130 write_unlock(&journal->j_state_lock); 1131 1132 out: 1133 return ret; 1134 } 1135 1136 /* 1137 * This is a variation of __jbd2_update_log_tail which checks for validity of 1138 * provided log tail and locks j_checkpoint_mutex. So it is safe against races 1139 * with other threads updating log tail. 1140 */ 1141 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1142 { 1143 mutex_lock_io(&journal->j_checkpoint_mutex); 1144 if (tid_gt(tid, journal->j_tail_sequence)) 1145 __jbd2_update_log_tail(journal, tid, block); 1146 mutex_unlock(&journal->j_checkpoint_mutex); 1147 } 1148 1149 struct jbd2_stats_proc_session { 1150 journal_t *journal; 1151 struct transaction_stats_s *stats; 1152 int start; 1153 int max; 1154 }; 1155 1156 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) 1157 { 1158 return *pos ? NULL : SEQ_START_TOKEN; 1159 } 1160 1161 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) 1162 { 1163 (*pos)++; 1164 return NULL; 1165 } 1166 1167 static int jbd2_seq_info_show(struct seq_file *seq, void *v) 1168 { 1169 struct jbd2_stats_proc_session *s = seq->private; 1170 1171 if (v != SEQ_START_TOKEN) 1172 return 0; 1173 seq_printf(seq, "%lu transactions (%lu requested), " 1174 "each up to %u blocks\n", 1175 s->stats->ts_tid, s->stats->ts_requested, 1176 s->journal->j_max_transaction_buffers); 1177 if (s->stats->ts_tid == 0) 1178 return 0; 1179 seq_printf(seq, "average: \n %ums waiting for transaction\n", 1180 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); 1181 seq_printf(seq, " %ums request delay\n", 1182 (s->stats->ts_requested == 0) ? 0 : 1183 jiffies_to_msecs(s->stats->run.rs_request_delay / 1184 s->stats->ts_requested)); 1185 seq_printf(seq, " %ums running transaction\n", 1186 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); 1187 seq_printf(seq, " %ums transaction was being locked\n", 1188 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); 1189 seq_printf(seq, " %ums flushing data (in ordered mode)\n", 1190 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); 1191 seq_printf(seq, " %ums logging transaction\n", 1192 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); 1193 seq_printf(seq, " %lluus average transaction commit time\n", 1194 div_u64(s->journal->j_average_commit_time, 1000)); 1195 seq_printf(seq, " %lu handles per transaction\n", 1196 s->stats->run.rs_handle_count / s->stats->ts_tid); 1197 seq_printf(seq, " %lu blocks per transaction\n", 1198 s->stats->run.rs_blocks / s->stats->ts_tid); 1199 seq_printf(seq, " %lu logged blocks per transaction\n", 1200 s->stats->run.rs_blocks_logged / s->stats->ts_tid); 1201 return 0; 1202 } 1203 1204 static void jbd2_seq_info_stop(struct seq_file *seq, void *v) 1205 { 1206 } 1207 1208 static const struct seq_operations jbd2_seq_info_ops = { 1209 .start = jbd2_seq_info_start, 1210 .next = jbd2_seq_info_next, 1211 .stop = jbd2_seq_info_stop, 1212 .show = jbd2_seq_info_show, 1213 }; 1214 1215 static int jbd2_seq_info_open(struct inode *inode, struct file *file) 1216 { 1217 journal_t *journal = pde_data(inode); 1218 struct jbd2_stats_proc_session *s; 1219 int rc, size; 1220 1221 s = kmalloc(sizeof(*s), GFP_KERNEL); 1222 if (s == NULL) 1223 return -ENOMEM; 1224 size = sizeof(struct transaction_stats_s); 1225 s->stats = kmalloc(size, GFP_KERNEL); 1226 if (s->stats == NULL) { 1227 kfree(s); 1228 return -ENOMEM; 1229 } 1230 spin_lock(&journal->j_history_lock); 1231 memcpy(s->stats, &journal->j_stats, size); 1232 s->journal = journal; 1233 spin_unlock(&journal->j_history_lock); 1234 1235 rc = seq_open(file, &jbd2_seq_info_ops); 1236 if (rc == 0) { 1237 struct seq_file *m = file->private_data; 1238 m->private = s; 1239 } else { 1240 kfree(s->stats); 1241 kfree(s); 1242 } 1243 return rc; 1244 1245 } 1246 1247 static int jbd2_seq_info_release(struct inode *inode, struct file *file) 1248 { 1249 struct seq_file *seq = file->private_data; 1250 struct jbd2_stats_proc_session *s = seq->private; 1251 kfree(s->stats); 1252 kfree(s); 1253 return seq_release(inode, file); 1254 } 1255 1256 static const struct proc_ops jbd2_info_proc_ops = { 1257 .proc_open = jbd2_seq_info_open, 1258 .proc_read = seq_read, 1259 .proc_lseek = seq_lseek, 1260 .proc_release = jbd2_seq_info_release, 1261 }; 1262 1263 static struct proc_dir_entry *proc_jbd2_stats; 1264 1265 static void jbd2_stats_proc_init(journal_t *journal) 1266 { 1267 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); 1268 if (journal->j_proc_entry) { 1269 proc_create_data("info", S_IRUGO, journal->j_proc_entry, 1270 &jbd2_info_proc_ops, journal); 1271 } 1272 } 1273 1274 static void jbd2_stats_proc_exit(journal_t *journal) 1275 { 1276 remove_proc_entry("info", journal->j_proc_entry); 1277 remove_proc_entry(journal->j_devname, proc_jbd2_stats); 1278 } 1279 1280 /* Minimum size of descriptor tag */ 1281 static int jbd2_min_tag_size(void) 1282 { 1283 /* 1284 * Tag with 32-bit block numbers does not use last four bytes of the 1285 * structure 1286 */ 1287 return sizeof(journal_block_tag_t) - 4; 1288 } 1289 1290 /** 1291 * jbd2_journal_shrink_scan() 1292 * @shrink: shrinker to work on 1293 * @sc: reclaim request to process 1294 * 1295 * Scan the checkpointed buffer on the checkpoint list and release the 1296 * journal_head. 1297 */ 1298 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink, 1299 struct shrink_control *sc) 1300 { 1301 journal_t *journal = container_of(shrink, journal_t, j_shrinker); 1302 unsigned long nr_to_scan = sc->nr_to_scan; 1303 unsigned long nr_shrunk; 1304 unsigned long count; 1305 1306 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1307 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count); 1308 1309 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan); 1310 1311 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1312 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count); 1313 1314 return nr_shrunk; 1315 } 1316 1317 /** 1318 * jbd2_journal_shrink_count() 1319 * @shrink: shrinker to work on 1320 * @sc: reclaim request to process 1321 * 1322 * Count the number of checkpoint buffers on the checkpoint list. 1323 */ 1324 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink, 1325 struct shrink_control *sc) 1326 { 1327 journal_t *journal = container_of(shrink, journal_t, j_shrinker); 1328 unsigned long count; 1329 1330 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1331 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count); 1332 1333 return count; 1334 } 1335 1336 /* 1337 * Management for journal control blocks: functions to create and 1338 * destroy journal_t structures, and to initialise and read existing 1339 * journal blocks from disk. */ 1340 1341 /* First: create and setup a journal_t object in memory. We initialise 1342 * very few fields yet: that has to wait until we have created the 1343 * journal structures from from scratch, or loaded them from disk. */ 1344 1345 static journal_t *journal_init_common(struct block_device *bdev, 1346 struct block_device *fs_dev, 1347 unsigned long long start, int len, int blocksize) 1348 { 1349 static struct lock_class_key jbd2_trans_commit_key; 1350 journal_t *journal; 1351 int err; 1352 struct buffer_head *bh; 1353 int n; 1354 1355 journal = kzalloc(sizeof(*journal), GFP_KERNEL); 1356 if (!journal) 1357 return NULL; 1358 1359 init_waitqueue_head(&journal->j_wait_transaction_locked); 1360 init_waitqueue_head(&journal->j_wait_done_commit); 1361 init_waitqueue_head(&journal->j_wait_commit); 1362 init_waitqueue_head(&journal->j_wait_updates); 1363 init_waitqueue_head(&journal->j_wait_reserved); 1364 init_waitqueue_head(&journal->j_fc_wait); 1365 mutex_init(&journal->j_abort_mutex); 1366 mutex_init(&journal->j_barrier); 1367 mutex_init(&journal->j_checkpoint_mutex); 1368 spin_lock_init(&journal->j_revoke_lock); 1369 spin_lock_init(&journal->j_list_lock); 1370 rwlock_init(&journal->j_state_lock); 1371 1372 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); 1373 journal->j_min_batch_time = 0; 1374 journal->j_max_batch_time = 15000; /* 15ms */ 1375 atomic_set(&journal->j_reserved_credits, 0); 1376 1377 /* The journal is marked for error until we succeed with recovery! */ 1378 journal->j_flags = JBD2_ABORT; 1379 1380 /* Set up a default-sized revoke table for the new mount. */ 1381 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); 1382 if (err) 1383 goto err_cleanup; 1384 1385 spin_lock_init(&journal->j_history_lock); 1386 1387 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle", 1388 &jbd2_trans_commit_key, 0); 1389 1390 /* journal descriptor can store up to n blocks -bzzz */ 1391 journal->j_blocksize = blocksize; 1392 journal->j_dev = bdev; 1393 journal->j_fs_dev = fs_dev; 1394 journal->j_blk_offset = start; 1395 journal->j_total_len = len; 1396 /* We need enough buffers to write out full descriptor block. */ 1397 n = journal->j_blocksize / jbd2_min_tag_size(); 1398 journal->j_wbufsize = n; 1399 journal->j_fc_wbuf = NULL; 1400 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *), 1401 GFP_KERNEL); 1402 if (!journal->j_wbuf) 1403 goto err_cleanup; 1404 1405 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize); 1406 if (!bh) { 1407 pr_err("%s: Cannot get buffer for journal superblock\n", 1408 __func__); 1409 goto err_cleanup; 1410 } 1411 journal->j_sb_buffer = bh; 1412 journal->j_superblock = (journal_superblock_t *)bh->b_data; 1413 1414 journal->j_shrink_transaction = NULL; 1415 journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan; 1416 journal->j_shrinker.count_objects = jbd2_journal_shrink_count; 1417 journal->j_shrinker.seeks = DEFAULT_SEEKS; 1418 journal->j_shrinker.batch = journal->j_max_transaction_buffers; 1419 1420 if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL)) 1421 goto err_cleanup; 1422 1423 if (register_shrinker(&journal->j_shrinker, "jbd2-journal:(%u:%u)", 1424 MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev))) { 1425 percpu_counter_destroy(&journal->j_checkpoint_jh_count); 1426 goto err_cleanup; 1427 } 1428 return journal; 1429 1430 err_cleanup: 1431 brelse(journal->j_sb_buffer); 1432 kfree(journal->j_wbuf); 1433 jbd2_journal_destroy_revoke(journal); 1434 kfree(journal); 1435 return NULL; 1436 } 1437 1438 /* jbd2_journal_init_dev and jbd2_journal_init_inode: 1439 * 1440 * Create a journal structure assigned some fixed set of disk blocks to 1441 * the journal. We don't actually touch those disk blocks yet, but we 1442 * need to set up all of the mapping information to tell the journaling 1443 * system where the journal blocks are. 1444 * 1445 */ 1446 1447 /** 1448 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure 1449 * @bdev: Block device on which to create the journal 1450 * @fs_dev: Device which hold journalled filesystem for this journal. 1451 * @start: Block nr Start of journal. 1452 * @len: Length of the journal in blocks. 1453 * @blocksize: blocksize of journalling device 1454 * 1455 * Returns: a newly created journal_t * 1456 * 1457 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous 1458 * range of blocks on an arbitrary block device. 1459 * 1460 */ 1461 journal_t *jbd2_journal_init_dev(struct block_device *bdev, 1462 struct block_device *fs_dev, 1463 unsigned long long start, int len, int blocksize) 1464 { 1465 journal_t *journal; 1466 1467 journal = journal_init_common(bdev, fs_dev, start, len, blocksize); 1468 if (!journal) 1469 return NULL; 1470 1471 snprintf(journal->j_devname, sizeof(journal->j_devname), 1472 "%pg", journal->j_dev); 1473 strreplace(journal->j_devname, '/', '!'); 1474 jbd2_stats_proc_init(journal); 1475 1476 return journal; 1477 } 1478 1479 /** 1480 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. 1481 * @inode: An inode to create the journal in 1482 * 1483 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as 1484 * the journal. The inode must exist already, must support bmap() and 1485 * must have all data blocks preallocated. 1486 */ 1487 journal_t *jbd2_journal_init_inode(struct inode *inode) 1488 { 1489 journal_t *journal; 1490 sector_t blocknr; 1491 int err = 0; 1492 1493 blocknr = 0; 1494 err = bmap(inode, &blocknr); 1495 1496 if (err || !blocknr) { 1497 pr_err("%s: Cannot locate journal superblock\n", 1498 __func__); 1499 return NULL; 1500 } 1501 1502 jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n", 1503 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size, 1504 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); 1505 1506 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev, 1507 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits, 1508 inode->i_sb->s_blocksize); 1509 if (!journal) 1510 return NULL; 1511 1512 journal->j_inode = inode; 1513 snprintf(journal->j_devname, sizeof(journal->j_devname), 1514 "%pg-%lu", journal->j_dev, journal->j_inode->i_ino); 1515 strreplace(journal->j_devname, '/', '!'); 1516 jbd2_stats_proc_init(journal); 1517 1518 return journal; 1519 } 1520 1521 /* 1522 * If the journal init or create aborts, we need to mark the journal 1523 * superblock as being NULL to prevent the journal destroy from writing 1524 * back a bogus superblock. 1525 */ 1526 static void journal_fail_superblock(journal_t *journal) 1527 { 1528 struct buffer_head *bh = journal->j_sb_buffer; 1529 brelse(bh); 1530 journal->j_sb_buffer = NULL; 1531 } 1532 1533 /* 1534 * Given a journal_t structure, initialise the various fields for 1535 * startup of a new journaling session. We use this both when creating 1536 * a journal, and after recovering an old journal to reset it for 1537 * subsequent use. 1538 */ 1539 1540 static int journal_reset(journal_t *journal) 1541 { 1542 journal_superblock_t *sb = journal->j_superblock; 1543 unsigned long long first, last; 1544 1545 first = be32_to_cpu(sb->s_first); 1546 last = be32_to_cpu(sb->s_maxlen); 1547 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { 1548 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", 1549 first, last); 1550 journal_fail_superblock(journal); 1551 return -EINVAL; 1552 } 1553 1554 journal->j_first = first; 1555 journal->j_last = last; 1556 1557 if (journal->j_head != 0 && journal->j_flags & JBD2_CYCLE_RECORD) { 1558 /* 1559 * Disable the cycled recording mode if the journal head block 1560 * number is not correct. 1561 */ 1562 if (journal->j_head < first || journal->j_head >= last) { 1563 printk(KERN_WARNING "JBD2: Incorrect Journal head block %lu, " 1564 "disable journal_cycle_record\n", 1565 journal->j_head); 1566 journal->j_head = journal->j_first; 1567 } 1568 } else { 1569 journal->j_head = journal->j_first; 1570 } 1571 journal->j_tail = journal->j_head; 1572 journal->j_free = journal->j_last - journal->j_first; 1573 1574 journal->j_tail_sequence = journal->j_transaction_sequence; 1575 journal->j_commit_sequence = journal->j_transaction_sequence - 1; 1576 journal->j_commit_request = journal->j_commit_sequence; 1577 1578 journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal); 1579 1580 /* 1581 * Now that journal recovery is done, turn fast commits off here. This 1582 * way, if fast commit was enabled before the crash but if now FS has 1583 * disabled it, we don't enable fast commits. 1584 */ 1585 jbd2_clear_feature_fast_commit(journal); 1586 1587 /* 1588 * As a special case, if the on-disk copy is already marked as needing 1589 * no recovery (s_start == 0), then we can safely defer the superblock 1590 * update until the next commit by setting JBD2_FLUSHED. This avoids 1591 * attempting a write to a potential-readonly device. 1592 */ 1593 if (sb->s_start == 0) { 1594 jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb " 1595 "(start %ld, seq %u, errno %d)\n", 1596 journal->j_tail, journal->j_tail_sequence, 1597 journal->j_errno); 1598 journal->j_flags |= JBD2_FLUSHED; 1599 } else { 1600 /* Lock here to make assertions happy... */ 1601 mutex_lock_io(&journal->j_checkpoint_mutex); 1602 /* 1603 * Update log tail information. We use REQ_FUA since new 1604 * transaction will start reusing journal space and so we 1605 * must make sure information about current log tail is on 1606 * disk before that. 1607 */ 1608 jbd2_journal_update_sb_log_tail(journal, 1609 journal->j_tail_sequence, 1610 journal->j_tail, 1611 REQ_SYNC | REQ_FUA); 1612 mutex_unlock(&journal->j_checkpoint_mutex); 1613 } 1614 return jbd2_journal_start_thread(journal); 1615 } 1616 1617 /* 1618 * This function expects that the caller will have locked the journal 1619 * buffer head, and will return with it unlocked 1620 */ 1621 static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags) 1622 { 1623 struct buffer_head *bh = journal->j_sb_buffer; 1624 journal_superblock_t *sb = journal->j_superblock; 1625 int ret = 0; 1626 1627 /* Buffer got discarded which means block device got invalidated */ 1628 if (!buffer_mapped(bh)) { 1629 unlock_buffer(bh); 1630 return -EIO; 1631 } 1632 1633 trace_jbd2_write_superblock(journal, write_flags); 1634 if (!(journal->j_flags & JBD2_BARRIER)) 1635 write_flags &= ~(REQ_FUA | REQ_PREFLUSH); 1636 if (buffer_write_io_error(bh)) { 1637 /* 1638 * Oh, dear. A previous attempt to write the journal 1639 * superblock failed. This could happen because the 1640 * USB device was yanked out. Or it could happen to 1641 * be a transient write error and maybe the block will 1642 * be remapped. Nothing we can do but to retry the 1643 * write and hope for the best. 1644 */ 1645 printk(KERN_ERR "JBD2: previous I/O error detected " 1646 "for journal superblock update for %s.\n", 1647 journal->j_devname); 1648 clear_buffer_write_io_error(bh); 1649 set_buffer_uptodate(bh); 1650 } 1651 if (jbd2_journal_has_csum_v2or3(journal)) 1652 sb->s_checksum = jbd2_superblock_csum(journal, sb); 1653 get_bh(bh); 1654 bh->b_end_io = end_buffer_write_sync; 1655 submit_bh(REQ_OP_WRITE | write_flags, bh); 1656 wait_on_buffer(bh); 1657 if (buffer_write_io_error(bh)) { 1658 clear_buffer_write_io_error(bh); 1659 set_buffer_uptodate(bh); 1660 ret = -EIO; 1661 } 1662 if (ret) { 1663 printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n", 1664 journal->j_devname); 1665 if (!is_journal_aborted(journal)) 1666 jbd2_journal_abort(journal, ret); 1667 } 1668 1669 return ret; 1670 } 1671 1672 /** 1673 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. 1674 * @journal: The journal to update. 1675 * @tail_tid: TID of the new transaction at the tail of the log 1676 * @tail_block: The first block of the transaction at the tail of the log 1677 * @write_flags: Flags for the journal sb write operation 1678 * 1679 * Update a journal's superblock information about log tail and write it to 1680 * disk, waiting for the IO to complete. 1681 */ 1682 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, 1683 unsigned long tail_block, 1684 blk_opf_t write_flags) 1685 { 1686 journal_superblock_t *sb = journal->j_superblock; 1687 int ret; 1688 1689 if (is_journal_aborted(journal)) 1690 return -EIO; 1691 if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) { 1692 jbd2_journal_abort(journal, -EIO); 1693 return -EIO; 1694 } 1695 1696 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1697 jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", 1698 tail_block, tail_tid); 1699 1700 lock_buffer(journal->j_sb_buffer); 1701 sb->s_sequence = cpu_to_be32(tail_tid); 1702 sb->s_start = cpu_to_be32(tail_block); 1703 1704 ret = jbd2_write_superblock(journal, write_flags); 1705 if (ret) 1706 goto out; 1707 1708 /* Log is no longer empty */ 1709 write_lock(&journal->j_state_lock); 1710 WARN_ON(!sb->s_sequence); 1711 journal->j_flags &= ~JBD2_FLUSHED; 1712 write_unlock(&journal->j_state_lock); 1713 1714 out: 1715 return ret; 1716 } 1717 1718 /** 1719 * jbd2_mark_journal_empty() - Mark on disk journal as empty. 1720 * @journal: The journal to update. 1721 * @write_flags: Flags for the journal sb write operation 1722 * 1723 * Update a journal's dynamic superblock fields to show that journal is empty. 1724 * Write updated superblock to disk waiting for IO to complete. 1725 */ 1726 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags) 1727 { 1728 journal_superblock_t *sb = journal->j_superblock; 1729 bool had_fast_commit = false; 1730 1731 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1732 lock_buffer(journal->j_sb_buffer); 1733 if (sb->s_start == 0) { /* Is it already empty? */ 1734 unlock_buffer(journal->j_sb_buffer); 1735 return; 1736 } 1737 1738 jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n", 1739 journal->j_tail_sequence); 1740 1741 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); 1742 sb->s_start = cpu_to_be32(0); 1743 sb->s_head = cpu_to_be32(journal->j_head); 1744 if (jbd2_has_feature_fast_commit(journal)) { 1745 /* 1746 * When journal is clean, no need to commit fast commit flag and 1747 * make file system incompatible with older kernels. 1748 */ 1749 jbd2_clear_feature_fast_commit(journal); 1750 had_fast_commit = true; 1751 } 1752 1753 jbd2_write_superblock(journal, write_flags); 1754 1755 if (had_fast_commit) 1756 jbd2_set_feature_fast_commit(journal); 1757 1758 /* Log is no longer empty */ 1759 write_lock(&journal->j_state_lock); 1760 journal->j_flags |= JBD2_FLUSHED; 1761 write_unlock(&journal->j_state_lock); 1762 } 1763 1764 /** 1765 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock) 1766 * @journal: The journal to erase. 1767 * @flags: A discard/zeroout request is sent for each physically contigous 1768 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or 1769 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation 1770 * to perform. 1771 * 1772 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes 1773 * will be explicitly written if no hardware offload is available, see 1774 * blkdev_issue_zeroout for more details. 1775 */ 1776 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags) 1777 { 1778 int err = 0; 1779 unsigned long block, log_offset; /* logical */ 1780 unsigned long long phys_block, block_start, block_stop; /* physical */ 1781 loff_t byte_start, byte_stop, byte_count; 1782 1783 /* flags must be set to either discard or zeroout */ 1784 if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags || 1785 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && 1786 (flags & JBD2_JOURNAL_FLUSH_ZEROOUT))) 1787 return -EINVAL; 1788 1789 if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && 1790 !bdev_max_discard_sectors(journal->j_dev)) 1791 return -EOPNOTSUPP; 1792 1793 /* 1794 * lookup block mapping and issue discard/zeroout for each 1795 * contiguous region 1796 */ 1797 log_offset = be32_to_cpu(journal->j_superblock->s_first); 1798 block_start = ~0ULL; 1799 for (block = log_offset; block < journal->j_total_len; block++) { 1800 err = jbd2_journal_bmap(journal, block, &phys_block); 1801 if (err) { 1802 pr_err("JBD2: bad block at offset %lu", block); 1803 return err; 1804 } 1805 1806 if (block_start == ~0ULL) { 1807 block_start = phys_block; 1808 block_stop = block_start - 1; 1809 } 1810 1811 /* 1812 * last block not contiguous with current block, 1813 * process last contiguous region and return to this block on 1814 * next loop 1815 */ 1816 if (phys_block != block_stop + 1) { 1817 block--; 1818 } else { 1819 block_stop++; 1820 /* 1821 * if this isn't the last block of journal, 1822 * no need to process now because next block may also 1823 * be part of this contiguous region 1824 */ 1825 if (block != journal->j_total_len - 1) 1826 continue; 1827 } 1828 1829 /* 1830 * end of contiguous region or this is last block of journal, 1831 * take care of the region 1832 */ 1833 byte_start = block_start * journal->j_blocksize; 1834 byte_stop = block_stop * journal->j_blocksize; 1835 byte_count = (block_stop - block_start + 1) * 1836 journal->j_blocksize; 1837 1838 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping, 1839 byte_start, byte_stop); 1840 1841 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) { 1842 err = blkdev_issue_discard(journal->j_dev, 1843 byte_start >> SECTOR_SHIFT, 1844 byte_count >> SECTOR_SHIFT, 1845 GFP_NOFS); 1846 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) { 1847 err = blkdev_issue_zeroout(journal->j_dev, 1848 byte_start >> SECTOR_SHIFT, 1849 byte_count >> SECTOR_SHIFT, 1850 GFP_NOFS, 0); 1851 } 1852 1853 if (unlikely(err != 0)) { 1854 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu", 1855 err, block_start, block_stop); 1856 return err; 1857 } 1858 1859 /* reset start and stop after processing a region */ 1860 block_start = ~0ULL; 1861 } 1862 1863 return blkdev_issue_flush(journal->j_dev); 1864 } 1865 1866 /** 1867 * jbd2_journal_update_sb_errno() - Update error in the journal. 1868 * @journal: The journal to update. 1869 * 1870 * Update a journal's errno. Write updated superblock to disk waiting for IO 1871 * to complete. 1872 */ 1873 void jbd2_journal_update_sb_errno(journal_t *journal) 1874 { 1875 journal_superblock_t *sb = journal->j_superblock; 1876 int errcode; 1877 1878 lock_buffer(journal->j_sb_buffer); 1879 errcode = journal->j_errno; 1880 if (errcode == -ESHUTDOWN) 1881 errcode = 0; 1882 jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode); 1883 sb->s_errno = cpu_to_be32(errcode); 1884 1885 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA); 1886 } 1887 EXPORT_SYMBOL(jbd2_journal_update_sb_errno); 1888 1889 static int journal_revoke_records_per_block(journal_t *journal) 1890 { 1891 int record_size; 1892 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t); 1893 1894 if (jbd2_has_feature_64bit(journal)) 1895 record_size = 8; 1896 else 1897 record_size = 4; 1898 1899 if (jbd2_journal_has_csum_v2or3(journal)) 1900 space -= sizeof(struct jbd2_journal_block_tail); 1901 return space / record_size; 1902 } 1903 1904 /* 1905 * Read the superblock for a given journal, performing initial 1906 * validation of the format. 1907 */ 1908 static int journal_get_superblock(journal_t *journal) 1909 { 1910 struct buffer_head *bh; 1911 journal_superblock_t *sb; 1912 int err; 1913 1914 bh = journal->j_sb_buffer; 1915 1916 J_ASSERT(bh != NULL); 1917 if (buffer_verified(bh)) 1918 return 0; 1919 1920 err = bh_read(bh, 0); 1921 if (err < 0) { 1922 printk(KERN_ERR 1923 "JBD2: IO error reading journal superblock\n"); 1924 goto out; 1925 } 1926 1927 sb = journal->j_superblock; 1928 1929 err = -EINVAL; 1930 1931 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || 1932 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { 1933 printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); 1934 goto out; 1935 } 1936 1937 if (be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V1 && 1938 be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V2) { 1939 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); 1940 goto out; 1941 } 1942 1943 if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) { 1944 printk(KERN_WARNING "JBD2: journal file too short\n"); 1945 goto out; 1946 } 1947 1948 if (be32_to_cpu(sb->s_first) == 0 || 1949 be32_to_cpu(sb->s_first) >= journal->j_total_len) { 1950 printk(KERN_WARNING 1951 "JBD2: Invalid start block of journal: %u\n", 1952 be32_to_cpu(sb->s_first)); 1953 goto out; 1954 } 1955 1956 if (jbd2_has_feature_csum2(journal) && 1957 jbd2_has_feature_csum3(journal)) { 1958 /* Can't have checksum v2 and v3 at the same time! */ 1959 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " 1960 "at the same time!\n"); 1961 goto out; 1962 } 1963 1964 if (jbd2_journal_has_csum_v2or3_feature(journal) && 1965 jbd2_has_feature_checksum(journal)) { 1966 /* Can't have checksum v1 and v2 on at the same time! */ 1967 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " 1968 "at the same time!\n"); 1969 goto out; 1970 } 1971 1972 if (!jbd2_verify_csum_type(journal, sb)) { 1973 printk(KERN_ERR "JBD2: Unknown checksum type\n"); 1974 goto out; 1975 } 1976 1977 /* Load the checksum driver */ 1978 if (jbd2_journal_has_csum_v2or3_feature(journal)) { 1979 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 1980 if (IS_ERR(journal->j_chksum_driver)) { 1981 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); 1982 err = PTR_ERR(journal->j_chksum_driver); 1983 journal->j_chksum_driver = NULL; 1984 goto out; 1985 } 1986 /* Check superblock checksum */ 1987 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) { 1988 printk(KERN_ERR "JBD2: journal checksum error\n"); 1989 err = -EFSBADCRC; 1990 goto out; 1991 } 1992 } 1993 set_buffer_verified(bh); 1994 return 0; 1995 1996 out: 1997 journal_fail_superblock(journal); 1998 return err; 1999 } 2000 2001 /* 2002 * Load the on-disk journal superblock and read the key fields into the 2003 * journal_t. 2004 */ 2005 2006 static int load_superblock(journal_t *journal) 2007 { 2008 int err; 2009 journal_superblock_t *sb; 2010 int num_fc_blocks; 2011 2012 err = journal_get_superblock(journal); 2013 if (err) 2014 return err; 2015 2016 sb = journal->j_superblock; 2017 2018 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); 2019 journal->j_tail = be32_to_cpu(sb->s_start); 2020 journal->j_first = be32_to_cpu(sb->s_first); 2021 journal->j_errno = be32_to_cpu(sb->s_errno); 2022 journal->j_last = be32_to_cpu(sb->s_maxlen); 2023 2024 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len) 2025 journal->j_total_len = be32_to_cpu(sb->s_maxlen); 2026 /* Precompute checksum seed for all metadata */ 2027 if (jbd2_journal_has_csum_v2or3(journal)) 2028 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, 2029 sizeof(sb->s_uuid)); 2030 journal->j_revoke_records_per_block = 2031 journal_revoke_records_per_block(journal); 2032 2033 if (jbd2_has_feature_fast_commit(journal)) { 2034 journal->j_fc_last = be32_to_cpu(sb->s_maxlen); 2035 num_fc_blocks = jbd2_journal_get_num_fc_blks(sb); 2036 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS) 2037 journal->j_last = journal->j_fc_last - num_fc_blocks; 2038 journal->j_fc_first = journal->j_last + 1; 2039 journal->j_fc_off = 0; 2040 } 2041 2042 return 0; 2043 } 2044 2045 2046 /** 2047 * jbd2_journal_load() - Read journal from disk. 2048 * @journal: Journal to act on. 2049 * 2050 * Given a journal_t structure which tells us which disk blocks contain 2051 * a journal, read the journal from disk to initialise the in-memory 2052 * structures. 2053 */ 2054 int jbd2_journal_load(journal_t *journal) 2055 { 2056 int err; 2057 journal_superblock_t *sb; 2058 2059 err = load_superblock(journal); 2060 if (err) 2061 return err; 2062 2063 sb = journal->j_superblock; 2064 2065 /* 2066 * If this is a V2 superblock, then we have to check the 2067 * features flags on it. 2068 */ 2069 if (jbd2_format_support_feature(journal)) { 2070 if ((sb->s_feature_ro_compat & 2071 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || 2072 (sb->s_feature_incompat & 2073 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { 2074 printk(KERN_WARNING 2075 "JBD2: Unrecognised features on journal\n"); 2076 return -EINVAL; 2077 } 2078 } 2079 2080 /* 2081 * Create a slab for this blocksize 2082 */ 2083 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); 2084 if (err) 2085 return err; 2086 2087 /* Let the recovery code check whether it needs to recover any 2088 * data from the journal. */ 2089 if (jbd2_journal_recover(journal)) 2090 goto recovery_error; 2091 2092 if (journal->j_failed_commit) { 2093 printk(KERN_ERR "JBD2: journal transaction %u on %s " 2094 "is corrupt.\n", journal->j_failed_commit, 2095 journal->j_devname); 2096 return -EFSCORRUPTED; 2097 } 2098 /* 2099 * clear JBD2_ABORT flag initialized in journal_init_common 2100 * here to update log tail information with the newest seq. 2101 */ 2102 journal->j_flags &= ~JBD2_ABORT; 2103 2104 /* OK, we've finished with the dynamic journal bits: 2105 * reinitialise the dynamic contents of the superblock in memory 2106 * and reset them on disk. */ 2107 if (journal_reset(journal)) 2108 goto recovery_error; 2109 2110 journal->j_flags |= JBD2_LOADED; 2111 return 0; 2112 2113 recovery_error: 2114 printk(KERN_WARNING "JBD2: recovery failed\n"); 2115 return -EIO; 2116 } 2117 2118 /** 2119 * jbd2_journal_destroy() - Release a journal_t structure. 2120 * @journal: Journal to act on. 2121 * 2122 * Release a journal_t structure once it is no longer in use by the 2123 * journaled object. 2124 * Return <0 if we couldn't clean up the journal. 2125 */ 2126 int jbd2_journal_destroy(journal_t *journal) 2127 { 2128 int err = 0; 2129 2130 /* Wait for the commit thread to wake up and die. */ 2131 journal_kill_thread(journal); 2132 2133 /* Force a final log commit */ 2134 if (journal->j_running_transaction) 2135 jbd2_journal_commit_transaction(journal); 2136 2137 /* Force any old transactions to disk */ 2138 2139 /* Totally anal locking here... */ 2140 spin_lock(&journal->j_list_lock); 2141 while (journal->j_checkpoint_transactions != NULL) { 2142 spin_unlock(&journal->j_list_lock); 2143 mutex_lock_io(&journal->j_checkpoint_mutex); 2144 err = jbd2_log_do_checkpoint(journal); 2145 mutex_unlock(&journal->j_checkpoint_mutex); 2146 /* 2147 * If checkpointing failed, just free the buffers to avoid 2148 * looping forever 2149 */ 2150 if (err) { 2151 jbd2_journal_destroy_checkpoint(journal); 2152 spin_lock(&journal->j_list_lock); 2153 break; 2154 } 2155 spin_lock(&journal->j_list_lock); 2156 } 2157 2158 J_ASSERT(journal->j_running_transaction == NULL); 2159 J_ASSERT(journal->j_committing_transaction == NULL); 2160 J_ASSERT(journal->j_checkpoint_transactions == NULL); 2161 spin_unlock(&journal->j_list_lock); 2162 2163 /* 2164 * OK, all checkpoint transactions have been checked, now check the 2165 * write out io error flag and abort the journal if some buffer failed 2166 * to write back to the original location, otherwise the filesystem 2167 * may become inconsistent. 2168 */ 2169 if (!is_journal_aborted(journal) && 2170 test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) 2171 jbd2_journal_abort(journal, -EIO); 2172 2173 if (journal->j_sb_buffer) { 2174 if (!is_journal_aborted(journal)) { 2175 mutex_lock_io(&journal->j_checkpoint_mutex); 2176 2177 write_lock(&journal->j_state_lock); 2178 journal->j_tail_sequence = 2179 ++journal->j_transaction_sequence; 2180 write_unlock(&journal->j_state_lock); 2181 2182 jbd2_mark_journal_empty(journal, 2183 REQ_SYNC | REQ_PREFLUSH | REQ_FUA); 2184 mutex_unlock(&journal->j_checkpoint_mutex); 2185 } else 2186 err = -EIO; 2187 brelse(journal->j_sb_buffer); 2188 } 2189 2190 if (journal->j_shrinker.flags & SHRINKER_REGISTERED) { 2191 percpu_counter_destroy(&journal->j_checkpoint_jh_count); 2192 unregister_shrinker(&journal->j_shrinker); 2193 } 2194 if (journal->j_proc_entry) 2195 jbd2_stats_proc_exit(journal); 2196 iput(journal->j_inode); 2197 if (journal->j_revoke) 2198 jbd2_journal_destroy_revoke(journal); 2199 if (journal->j_chksum_driver) 2200 crypto_free_shash(journal->j_chksum_driver); 2201 kfree(journal->j_fc_wbuf); 2202 kfree(journal->j_wbuf); 2203 kfree(journal); 2204 2205 return err; 2206 } 2207 2208 2209 /** 2210 * jbd2_journal_check_used_features() - Check if features specified are used. 2211 * @journal: Journal to check. 2212 * @compat: bitmask of compatible features 2213 * @ro: bitmask of features that force read-only mount 2214 * @incompat: bitmask of incompatible features 2215 * 2216 * Check whether the journal uses all of a given set of 2217 * features. Return true (non-zero) if it does. 2218 **/ 2219 2220 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat, 2221 unsigned long ro, unsigned long incompat) 2222 { 2223 journal_superblock_t *sb; 2224 2225 if (!compat && !ro && !incompat) 2226 return 1; 2227 if (journal_get_superblock(journal)) 2228 return 0; 2229 if (!jbd2_format_support_feature(journal)) 2230 return 0; 2231 2232 sb = journal->j_superblock; 2233 2234 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && 2235 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && 2236 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) 2237 return 1; 2238 2239 return 0; 2240 } 2241 2242 /** 2243 * jbd2_journal_check_available_features() - Check feature set in journalling layer 2244 * @journal: Journal to check. 2245 * @compat: bitmask of compatible features 2246 * @ro: bitmask of features that force read-only mount 2247 * @incompat: bitmask of incompatible features 2248 * 2249 * Check whether the journaling code supports the use of 2250 * all of a given set of features on this journal. Return true 2251 * (non-zero) if it can. */ 2252 2253 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat, 2254 unsigned long ro, unsigned long incompat) 2255 { 2256 if (!compat && !ro && !incompat) 2257 return 1; 2258 2259 if (!jbd2_format_support_feature(journal)) 2260 return 0; 2261 2262 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && 2263 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && 2264 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) 2265 return 1; 2266 2267 return 0; 2268 } 2269 2270 static int 2271 jbd2_journal_initialize_fast_commit(journal_t *journal) 2272 { 2273 journal_superblock_t *sb = journal->j_superblock; 2274 unsigned long long num_fc_blks; 2275 2276 num_fc_blks = jbd2_journal_get_num_fc_blks(sb); 2277 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS) 2278 return -ENOSPC; 2279 2280 /* Are we called twice? */ 2281 WARN_ON(journal->j_fc_wbuf != NULL); 2282 journal->j_fc_wbuf = kmalloc_array(num_fc_blks, 2283 sizeof(struct buffer_head *), GFP_KERNEL); 2284 if (!journal->j_fc_wbuf) 2285 return -ENOMEM; 2286 2287 journal->j_fc_wbufsize = num_fc_blks; 2288 journal->j_fc_last = journal->j_last; 2289 journal->j_last = journal->j_fc_last - num_fc_blks; 2290 journal->j_fc_first = journal->j_last + 1; 2291 journal->j_fc_off = 0; 2292 journal->j_free = journal->j_last - journal->j_first; 2293 journal->j_max_transaction_buffers = 2294 jbd2_journal_get_max_txn_bufs(journal); 2295 2296 return 0; 2297 } 2298 2299 /** 2300 * jbd2_journal_set_features() - Mark a given journal feature in the superblock 2301 * @journal: Journal to act on. 2302 * @compat: bitmask of compatible features 2303 * @ro: bitmask of features that force read-only mount 2304 * @incompat: bitmask of incompatible features 2305 * 2306 * Mark a given journal feature as present on the 2307 * superblock. Returns true if the requested features could be set. 2308 * 2309 */ 2310 2311 int jbd2_journal_set_features(journal_t *journal, unsigned long compat, 2312 unsigned long ro, unsigned long incompat) 2313 { 2314 #define INCOMPAT_FEATURE_ON(f) \ 2315 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) 2316 #define COMPAT_FEATURE_ON(f) \ 2317 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) 2318 journal_superblock_t *sb; 2319 2320 if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) 2321 return 1; 2322 2323 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) 2324 return 0; 2325 2326 /* If enabling v2 checksums, turn on v3 instead */ 2327 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { 2328 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; 2329 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; 2330 } 2331 2332 /* Asking for checksumming v3 and v1? Only give them v3. */ 2333 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && 2334 compat & JBD2_FEATURE_COMPAT_CHECKSUM) 2335 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; 2336 2337 jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", 2338 compat, ro, incompat); 2339 2340 sb = journal->j_superblock; 2341 2342 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) { 2343 if (jbd2_journal_initialize_fast_commit(journal)) { 2344 pr_err("JBD2: Cannot enable fast commits.\n"); 2345 return 0; 2346 } 2347 } 2348 2349 /* Load the checksum driver if necessary */ 2350 if ((journal->j_chksum_driver == NULL) && 2351 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 2352 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 2353 if (IS_ERR(journal->j_chksum_driver)) { 2354 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); 2355 journal->j_chksum_driver = NULL; 2356 return 0; 2357 } 2358 /* Precompute checksum seed for all metadata */ 2359 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, 2360 sizeof(sb->s_uuid)); 2361 } 2362 2363 lock_buffer(journal->j_sb_buffer); 2364 2365 /* If enabling v3 checksums, update superblock */ 2366 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 2367 sb->s_checksum_type = JBD2_CRC32C_CHKSUM; 2368 sb->s_feature_compat &= 2369 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); 2370 } 2371 2372 /* If enabling v1 checksums, downgrade superblock */ 2373 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) 2374 sb->s_feature_incompat &= 2375 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | 2376 JBD2_FEATURE_INCOMPAT_CSUM_V3); 2377 2378 sb->s_feature_compat |= cpu_to_be32(compat); 2379 sb->s_feature_ro_compat |= cpu_to_be32(ro); 2380 sb->s_feature_incompat |= cpu_to_be32(incompat); 2381 unlock_buffer(journal->j_sb_buffer); 2382 journal->j_revoke_records_per_block = 2383 journal_revoke_records_per_block(journal); 2384 2385 return 1; 2386 #undef COMPAT_FEATURE_ON 2387 #undef INCOMPAT_FEATURE_ON 2388 } 2389 2390 /* 2391 * jbd2_journal_clear_features() - Clear a given journal feature in the 2392 * superblock 2393 * @journal: Journal to act on. 2394 * @compat: bitmask of compatible features 2395 * @ro: bitmask of features that force read-only mount 2396 * @incompat: bitmask of incompatible features 2397 * 2398 * Clear a given journal feature as present on the 2399 * superblock. 2400 */ 2401 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, 2402 unsigned long ro, unsigned long incompat) 2403 { 2404 journal_superblock_t *sb; 2405 2406 jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", 2407 compat, ro, incompat); 2408 2409 sb = journal->j_superblock; 2410 2411 sb->s_feature_compat &= ~cpu_to_be32(compat); 2412 sb->s_feature_ro_compat &= ~cpu_to_be32(ro); 2413 sb->s_feature_incompat &= ~cpu_to_be32(incompat); 2414 journal->j_revoke_records_per_block = 2415 journal_revoke_records_per_block(journal); 2416 } 2417 EXPORT_SYMBOL(jbd2_journal_clear_features); 2418 2419 /** 2420 * jbd2_journal_flush() - Flush journal 2421 * @journal: Journal to act on. 2422 * @flags: optional operation on the journal blocks after the flush (see below) 2423 * 2424 * Flush all data for a given journal to disk and empty the journal. 2425 * Filesystems can use this when remounting readonly to ensure that 2426 * recovery does not need to happen on remount. Optionally, a discard or zeroout 2427 * can be issued on the journal blocks after flushing. 2428 * 2429 * flags: 2430 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks 2431 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks 2432 */ 2433 int jbd2_journal_flush(journal_t *journal, unsigned int flags) 2434 { 2435 int err = 0; 2436 transaction_t *transaction = NULL; 2437 2438 write_lock(&journal->j_state_lock); 2439 2440 /* Force everything buffered to the log... */ 2441 if (journal->j_running_transaction) { 2442 transaction = journal->j_running_transaction; 2443 __jbd2_log_start_commit(journal, transaction->t_tid); 2444 } else if (journal->j_committing_transaction) 2445 transaction = journal->j_committing_transaction; 2446 2447 /* Wait for the log commit to complete... */ 2448 if (transaction) { 2449 tid_t tid = transaction->t_tid; 2450 2451 write_unlock(&journal->j_state_lock); 2452 jbd2_log_wait_commit(journal, tid); 2453 } else { 2454 write_unlock(&journal->j_state_lock); 2455 } 2456 2457 /* ...and flush everything in the log out to disk. */ 2458 spin_lock(&journal->j_list_lock); 2459 while (!err && journal->j_checkpoint_transactions != NULL) { 2460 spin_unlock(&journal->j_list_lock); 2461 mutex_lock_io(&journal->j_checkpoint_mutex); 2462 err = jbd2_log_do_checkpoint(journal); 2463 mutex_unlock(&journal->j_checkpoint_mutex); 2464 spin_lock(&journal->j_list_lock); 2465 } 2466 spin_unlock(&journal->j_list_lock); 2467 2468 if (is_journal_aborted(journal)) 2469 return -EIO; 2470 2471 mutex_lock_io(&journal->j_checkpoint_mutex); 2472 if (!err) { 2473 err = jbd2_cleanup_journal_tail(journal); 2474 if (err < 0) { 2475 mutex_unlock(&journal->j_checkpoint_mutex); 2476 goto out; 2477 } 2478 err = 0; 2479 } 2480 2481 /* Finally, mark the journal as really needing no recovery. 2482 * This sets s_start==0 in the underlying superblock, which is 2483 * the magic code for a fully-recovered superblock. Any future 2484 * commits of data to the journal will restore the current 2485 * s_start value. */ 2486 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); 2487 2488 if (flags) 2489 err = __jbd2_journal_erase(journal, flags); 2490 2491 mutex_unlock(&journal->j_checkpoint_mutex); 2492 write_lock(&journal->j_state_lock); 2493 J_ASSERT(!journal->j_running_transaction); 2494 J_ASSERT(!journal->j_committing_transaction); 2495 J_ASSERT(!journal->j_checkpoint_transactions); 2496 J_ASSERT(journal->j_head == journal->j_tail); 2497 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); 2498 write_unlock(&journal->j_state_lock); 2499 out: 2500 return err; 2501 } 2502 2503 /** 2504 * jbd2_journal_wipe() - Wipe journal contents 2505 * @journal: Journal to act on. 2506 * @write: flag (see below) 2507 * 2508 * Wipe out all of the contents of a journal, safely. This will produce 2509 * a warning if the journal contains any valid recovery information. 2510 * Must be called between journal_init_*() and jbd2_journal_load(). 2511 * 2512 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise 2513 * we merely suppress recovery. 2514 */ 2515 2516 int jbd2_journal_wipe(journal_t *journal, int write) 2517 { 2518 int err = 0; 2519 2520 J_ASSERT (!(journal->j_flags & JBD2_LOADED)); 2521 2522 err = load_superblock(journal); 2523 if (err) 2524 return err; 2525 2526 if (!journal->j_tail) 2527 goto no_recovery; 2528 2529 printk(KERN_WARNING "JBD2: %s recovery information on journal\n", 2530 write ? "Clearing" : "Ignoring"); 2531 2532 err = jbd2_journal_skip_recovery(journal); 2533 if (write) { 2534 /* Lock to make assertions happy... */ 2535 mutex_lock_io(&journal->j_checkpoint_mutex); 2536 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); 2537 mutex_unlock(&journal->j_checkpoint_mutex); 2538 } 2539 2540 no_recovery: 2541 return err; 2542 } 2543 2544 /** 2545 * jbd2_journal_abort () - Shutdown the journal immediately. 2546 * @journal: the journal to shutdown. 2547 * @errno: an error number to record in the journal indicating 2548 * the reason for the shutdown. 2549 * 2550 * Perform a complete, immediate shutdown of the ENTIRE 2551 * journal (not of a single transaction). This operation cannot be 2552 * undone without closing and reopening the journal. 2553 * 2554 * The jbd2_journal_abort function is intended to support higher level error 2555 * recovery mechanisms such as the ext2/ext3 remount-readonly error 2556 * mode. 2557 * 2558 * Journal abort has very specific semantics. Any existing dirty, 2559 * unjournaled buffers in the main filesystem will still be written to 2560 * disk by bdflush, but the journaling mechanism will be suspended 2561 * immediately and no further transaction commits will be honoured. 2562 * 2563 * Any dirty, journaled buffers will be written back to disk without 2564 * hitting the journal. Atomicity cannot be guaranteed on an aborted 2565 * filesystem, but we _do_ attempt to leave as much data as possible 2566 * behind for fsck to use for cleanup. 2567 * 2568 * Any attempt to get a new transaction handle on a journal which is in 2569 * ABORT state will just result in an -EROFS error return. A 2570 * jbd2_journal_stop on an existing handle will return -EIO if we have 2571 * entered abort state during the update. 2572 * 2573 * Recursive transactions are not disturbed by journal abort until the 2574 * final jbd2_journal_stop, which will receive the -EIO error. 2575 * 2576 * Finally, the jbd2_journal_abort call allows the caller to supply an errno 2577 * which will be recorded (if possible) in the journal superblock. This 2578 * allows a client to record failure conditions in the middle of a 2579 * transaction without having to complete the transaction to record the 2580 * failure to disk. ext3_error, for example, now uses this 2581 * functionality. 2582 * 2583 */ 2584 2585 void jbd2_journal_abort(journal_t *journal, int errno) 2586 { 2587 transaction_t *transaction; 2588 2589 /* 2590 * Lock the aborting procedure until everything is done, this avoid 2591 * races between filesystem's error handling flow (e.g. ext4_abort()), 2592 * ensure panic after the error info is written into journal's 2593 * superblock. 2594 */ 2595 mutex_lock(&journal->j_abort_mutex); 2596 /* 2597 * ESHUTDOWN always takes precedence because a file system check 2598 * caused by any other journal abort error is not required after 2599 * a shutdown triggered. 2600 */ 2601 write_lock(&journal->j_state_lock); 2602 if (journal->j_flags & JBD2_ABORT) { 2603 int old_errno = journal->j_errno; 2604 2605 write_unlock(&journal->j_state_lock); 2606 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) { 2607 journal->j_errno = errno; 2608 jbd2_journal_update_sb_errno(journal); 2609 } 2610 mutex_unlock(&journal->j_abort_mutex); 2611 return; 2612 } 2613 2614 /* 2615 * Mark the abort as occurred and start current running transaction 2616 * to release all journaled buffer. 2617 */ 2618 pr_err("Aborting journal on device %s.\n", journal->j_devname); 2619 2620 journal->j_flags |= JBD2_ABORT; 2621 journal->j_errno = errno; 2622 transaction = journal->j_running_transaction; 2623 if (transaction) 2624 __jbd2_log_start_commit(journal, transaction->t_tid); 2625 write_unlock(&journal->j_state_lock); 2626 2627 /* 2628 * Record errno to the journal super block, so that fsck and jbd2 2629 * layer could realise that a filesystem check is needed. 2630 */ 2631 jbd2_journal_update_sb_errno(journal); 2632 mutex_unlock(&journal->j_abort_mutex); 2633 } 2634 2635 /** 2636 * jbd2_journal_errno() - returns the journal's error state. 2637 * @journal: journal to examine. 2638 * 2639 * This is the errno number set with jbd2_journal_abort(), the last 2640 * time the journal was mounted - if the journal was stopped 2641 * without calling abort this will be 0. 2642 * 2643 * If the journal has been aborted on this mount time -EROFS will 2644 * be returned. 2645 */ 2646 int jbd2_journal_errno(journal_t *journal) 2647 { 2648 int err; 2649 2650 read_lock(&journal->j_state_lock); 2651 if (journal->j_flags & JBD2_ABORT) 2652 err = -EROFS; 2653 else 2654 err = journal->j_errno; 2655 read_unlock(&journal->j_state_lock); 2656 return err; 2657 } 2658 2659 /** 2660 * jbd2_journal_clear_err() - clears the journal's error state 2661 * @journal: journal to act on. 2662 * 2663 * An error must be cleared or acked to take a FS out of readonly 2664 * mode. 2665 */ 2666 int jbd2_journal_clear_err(journal_t *journal) 2667 { 2668 int err = 0; 2669 2670 write_lock(&journal->j_state_lock); 2671 if (journal->j_flags & JBD2_ABORT) 2672 err = -EROFS; 2673 else 2674 journal->j_errno = 0; 2675 write_unlock(&journal->j_state_lock); 2676 return err; 2677 } 2678 2679 /** 2680 * jbd2_journal_ack_err() - Ack journal err. 2681 * @journal: journal to act on. 2682 * 2683 * An error must be cleared or acked to take a FS out of readonly 2684 * mode. 2685 */ 2686 void jbd2_journal_ack_err(journal_t *journal) 2687 { 2688 write_lock(&journal->j_state_lock); 2689 if (journal->j_errno) 2690 journal->j_flags |= JBD2_ACK_ERR; 2691 write_unlock(&journal->j_state_lock); 2692 } 2693 2694 int jbd2_journal_blocks_per_page(struct inode *inode) 2695 { 2696 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); 2697 } 2698 2699 /* 2700 * helper functions to deal with 32 or 64bit block numbers. 2701 */ 2702 size_t journal_tag_bytes(journal_t *journal) 2703 { 2704 size_t sz; 2705 2706 if (jbd2_has_feature_csum3(journal)) 2707 return sizeof(journal_block_tag3_t); 2708 2709 sz = sizeof(journal_block_tag_t); 2710 2711 if (jbd2_has_feature_csum2(journal)) 2712 sz += sizeof(__u16); 2713 2714 if (jbd2_has_feature_64bit(journal)) 2715 return sz; 2716 else 2717 return sz - sizeof(__u32); 2718 } 2719 2720 /* 2721 * JBD memory management 2722 * 2723 * These functions are used to allocate block-sized chunks of memory 2724 * used for making copies of buffer_head data. Very often it will be 2725 * page-sized chunks of data, but sometimes it will be in 2726 * sub-page-size chunks. (For example, 16k pages on Power systems 2727 * with a 4k block file system.) For blocks smaller than a page, we 2728 * use a SLAB allocator. There are slab caches for each block size, 2729 * which are allocated at mount time, if necessary, and we only free 2730 * (all of) the slab caches when/if the jbd2 module is unloaded. For 2731 * this reason we don't need to a mutex to protect access to 2732 * jbd2_slab[] allocating or releasing memory; only in 2733 * jbd2_journal_create_slab(). 2734 */ 2735 #define JBD2_MAX_SLABS 8 2736 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; 2737 2738 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { 2739 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", 2740 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" 2741 }; 2742 2743 2744 static void jbd2_journal_destroy_slabs(void) 2745 { 2746 int i; 2747 2748 for (i = 0; i < JBD2_MAX_SLABS; i++) { 2749 kmem_cache_destroy(jbd2_slab[i]); 2750 jbd2_slab[i] = NULL; 2751 } 2752 } 2753 2754 static int jbd2_journal_create_slab(size_t size) 2755 { 2756 static DEFINE_MUTEX(jbd2_slab_create_mutex); 2757 int i = order_base_2(size) - 10; 2758 size_t slab_size; 2759 2760 if (size == PAGE_SIZE) 2761 return 0; 2762 2763 if (i >= JBD2_MAX_SLABS) 2764 return -EINVAL; 2765 2766 if (unlikely(i < 0)) 2767 i = 0; 2768 mutex_lock(&jbd2_slab_create_mutex); 2769 if (jbd2_slab[i]) { 2770 mutex_unlock(&jbd2_slab_create_mutex); 2771 return 0; /* Already created */ 2772 } 2773 2774 slab_size = 1 << (i+10); 2775 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, 2776 slab_size, 0, NULL); 2777 mutex_unlock(&jbd2_slab_create_mutex); 2778 if (!jbd2_slab[i]) { 2779 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); 2780 return -ENOMEM; 2781 } 2782 return 0; 2783 } 2784 2785 static struct kmem_cache *get_slab(size_t size) 2786 { 2787 int i = order_base_2(size) - 10; 2788 2789 BUG_ON(i >= JBD2_MAX_SLABS); 2790 if (unlikely(i < 0)) 2791 i = 0; 2792 BUG_ON(jbd2_slab[i] == NULL); 2793 return jbd2_slab[i]; 2794 } 2795 2796 void *jbd2_alloc(size_t size, gfp_t flags) 2797 { 2798 void *ptr; 2799 2800 BUG_ON(size & (size-1)); /* Must be a power of 2 */ 2801 2802 if (size < PAGE_SIZE) 2803 ptr = kmem_cache_alloc(get_slab(size), flags); 2804 else 2805 ptr = (void *)__get_free_pages(flags, get_order(size)); 2806 2807 /* Check alignment; SLUB has gotten this wrong in the past, 2808 * and this can lead to user data corruption! */ 2809 BUG_ON(((unsigned long) ptr) & (size-1)); 2810 2811 return ptr; 2812 } 2813 2814 void jbd2_free(void *ptr, size_t size) 2815 { 2816 if (size < PAGE_SIZE) 2817 kmem_cache_free(get_slab(size), ptr); 2818 else 2819 free_pages((unsigned long)ptr, get_order(size)); 2820 }; 2821 2822 /* 2823 * Journal_head storage management 2824 */ 2825 static struct kmem_cache *jbd2_journal_head_cache; 2826 #ifdef CONFIG_JBD2_DEBUG 2827 static atomic_t nr_journal_heads = ATOMIC_INIT(0); 2828 #endif 2829 2830 static int __init jbd2_journal_init_journal_head_cache(void) 2831 { 2832 J_ASSERT(!jbd2_journal_head_cache); 2833 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", 2834 sizeof(struct journal_head), 2835 0, /* offset */ 2836 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU, 2837 NULL); /* ctor */ 2838 if (!jbd2_journal_head_cache) { 2839 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); 2840 return -ENOMEM; 2841 } 2842 return 0; 2843 } 2844 2845 static void jbd2_journal_destroy_journal_head_cache(void) 2846 { 2847 kmem_cache_destroy(jbd2_journal_head_cache); 2848 jbd2_journal_head_cache = NULL; 2849 } 2850 2851 /* 2852 * journal_head splicing and dicing 2853 */ 2854 static struct journal_head *journal_alloc_journal_head(void) 2855 { 2856 struct journal_head *ret; 2857 2858 #ifdef CONFIG_JBD2_DEBUG 2859 atomic_inc(&nr_journal_heads); 2860 #endif 2861 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); 2862 if (!ret) { 2863 jbd2_debug(1, "out of memory for journal_head\n"); 2864 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); 2865 ret = kmem_cache_zalloc(jbd2_journal_head_cache, 2866 GFP_NOFS | __GFP_NOFAIL); 2867 } 2868 if (ret) 2869 spin_lock_init(&ret->b_state_lock); 2870 return ret; 2871 } 2872 2873 static void journal_free_journal_head(struct journal_head *jh) 2874 { 2875 #ifdef CONFIG_JBD2_DEBUG 2876 atomic_dec(&nr_journal_heads); 2877 memset(jh, JBD2_POISON_FREE, sizeof(*jh)); 2878 #endif 2879 kmem_cache_free(jbd2_journal_head_cache, jh); 2880 } 2881 2882 /* 2883 * A journal_head is attached to a buffer_head whenever JBD has an 2884 * interest in the buffer. 2885 * 2886 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit 2887 * is set. This bit is tested in core kernel code where we need to take 2888 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable 2889 * there. 2890 * 2891 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. 2892 * 2893 * When a buffer has its BH_JBD bit set it is immune from being released by 2894 * core kernel code, mainly via ->b_count. 2895 * 2896 * A journal_head is detached from its buffer_head when the journal_head's 2897 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint 2898 * transaction (b_cp_transaction) hold their references to b_jcount. 2899 * 2900 * Various places in the kernel want to attach a journal_head to a buffer_head 2901 * _before_ attaching the journal_head to a transaction. To protect the 2902 * journal_head in this situation, jbd2_journal_add_journal_head elevates the 2903 * journal_head's b_jcount refcount by one. The caller must call 2904 * jbd2_journal_put_journal_head() to undo this. 2905 * 2906 * So the typical usage would be: 2907 * 2908 * (Attach a journal_head if needed. Increments b_jcount) 2909 * struct journal_head *jh = jbd2_journal_add_journal_head(bh); 2910 * ... 2911 * (Get another reference for transaction) 2912 * jbd2_journal_grab_journal_head(bh); 2913 * jh->b_transaction = xxx; 2914 * (Put original reference) 2915 * jbd2_journal_put_journal_head(jh); 2916 */ 2917 2918 /* 2919 * Give a buffer_head a journal_head. 2920 * 2921 * May sleep. 2922 */ 2923 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) 2924 { 2925 struct journal_head *jh; 2926 struct journal_head *new_jh = NULL; 2927 2928 repeat: 2929 if (!buffer_jbd(bh)) 2930 new_jh = journal_alloc_journal_head(); 2931 2932 jbd_lock_bh_journal_head(bh); 2933 if (buffer_jbd(bh)) { 2934 jh = bh2jh(bh); 2935 } else { 2936 J_ASSERT_BH(bh, 2937 (atomic_read(&bh->b_count) > 0) || 2938 (bh->b_folio && bh->b_folio->mapping)); 2939 2940 if (!new_jh) { 2941 jbd_unlock_bh_journal_head(bh); 2942 goto repeat; 2943 } 2944 2945 jh = new_jh; 2946 new_jh = NULL; /* We consumed it */ 2947 set_buffer_jbd(bh); 2948 bh->b_private = jh; 2949 jh->b_bh = bh; 2950 get_bh(bh); 2951 BUFFER_TRACE(bh, "added journal_head"); 2952 } 2953 jh->b_jcount++; 2954 jbd_unlock_bh_journal_head(bh); 2955 if (new_jh) 2956 journal_free_journal_head(new_jh); 2957 return bh->b_private; 2958 } 2959 2960 /* 2961 * Grab a ref against this buffer_head's journal_head. If it ended up not 2962 * having a journal_head, return NULL 2963 */ 2964 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) 2965 { 2966 struct journal_head *jh = NULL; 2967 2968 jbd_lock_bh_journal_head(bh); 2969 if (buffer_jbd(bh)) { 2970 jh = bh2jh(bh); 2971 jh->b_jcount++; 2972 } 2973 jbd_unlock_bh_journal_head(bh); 2974 return jh; 2975 } 2976 EXPORT_SYMBOL(jbd2_journal_grab_journal_head); 2977 2978 static void __journal_remove_journal_head(struct buffer_head *bh) 2979 { 2980 struct journal_head *jh = bh2jh(bh); 2981 2982 J_ASSERT_JH(jh, jh->b_transaction == NULL); 2983 J_ASSERT_JH(jh, jh->b_next_transaction == NULL); 2984 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); 2985 J_ASSERT_JH(jh, jh->b_jlist == BJ_None); 2986 J_ASSERT_BH(bh, buffer_jbd(bh)); 2987 J_ASSERT_BH(bh, jh2bh(jh) == bh); 2988 BUFFER_TRACE(bh, "remove journal_head"); 2989 2990 /* Unlink before dropping the lock */ 2991 bh->b_private = NULL; 2992 jh->b_bh = NULL; /* debug, really */ 2993 clear_buffer_jbd(bh); 2994 } 2995 2996 static void journal_release_journal_head(struct journal_head *jh, size_t b_size) 2997 { 2998 if (jh->b_frozen_data) { 2999 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); 3000 jbd2_free(jh->b_frozen_data, b_size); 3001 } 3002 if (jh->b_committed_data) { 3003 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); 3004 jbd2_free(jh->b_committed_data, b_size); 3005 } 3006 journal_free_journal_head(jh); 3007 } 3008 3009 /* 3010 * Drop a reference on the passed journal_head. If it fell to zero then 3011 * release the journal_head from the buffer_head. 3012 */ 3013 void jbd2_journal_put_journal_head(struct journal_head *jh) 3014 { 3015 struct buffer_head *bh = jh2bh(jh); 3016 3017 jbd_lock_bh_journal_head(bh); 3018 J_ASSERT_JH(jh, jh->b_jcount > 0); 3019 --jh->b_jcount; 3020 if (!jh->b_jcount) { 3021 __journal_remove_journal_head(bh); 3022 jbd_unlock_bh_journal_head(bh); 3023 journal_release_journal_head(jh, bh->b_size); 3024 __brelse(bh); 3025 } else { 3026 jbd_unlock_bh_journal_head(bh); 3027 } 3028 } 3029 EXPORT_SYMBOL(jbd2_journal_put_journal_head); 3030 3031 /* 3032 * Initialize jbd inode head 3033 */ 3034 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) 3035 { 3036 jinode->i_transaction = NULL; 3037 jinode->i_next_transaction = NULL; 3038 jinode->i_vfs_inode = inode; 3039 jinode->i_flags = 0; 3040 jinode->i_dirty_start = 0; 3041 jinode->i_dirty_end = 0; 3042 INIT_LIST_HEAD(&jinode->i_list); 3043 } 3044 3045 /* 3046 * Function to be called before we start removing inode from memory (i.e., 3047 * clear_inode() is a fine place to be called from). It removes inode from 3048 * transaction's lists. 3049 */ 3050 void jbd2_journal_release_jbd_inode(journal_t *journal, 3051 struct jbd2_inode *jinode) 3052 { 3053 if (!journal) 3054 return; 3055 restart: 3056 spin_lock(&journal->j_list_lock); 3057 /* Is commit writing out inode - we have to wait */ 3058 if (jinode->i_flags & JI_COMMIT_RUNNING) { 3059 wait_queue_head_t *wq; 3060 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); 3061 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); 3062 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 3063 spin_unlock(&journal->j_list_lock); 3064 schedule(); 3065 finish_wait(wq, &wait.wq_entry); 3066 goto restart; 3067 } 3068 3069 if (jinode->i_transaction) { 3070 list_del(&jinode->i_list); 3071 jinode->i_transaction = NULL; 3072 } 3073 spin_unlock(&journal->j_list_lock); 3074 } 3075 3076 3077 #ifdef CONFIG_PROC_FS 3078 3079 #define JBD2_STATS_PROC_NAME "fs/jbd2" 3080 3081 static void __init jbd2_create_jbd_stats_proc_entry(void) 3082 { 3083 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); 3084 } 3085 3086 static void __exit jbd2_remove_jbd_stats_proc_entry(void) 3087 { 3088 if (proc_jbd2_stats) 3089 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); 3090 } 3091 3092 #else 3093 3094 #define jbd2_create_jbd_stats_proc_entry() do {} while (0) 3095 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) 3096 3097 #endif 3098 3099 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; 3100 3101 static int __init jbd2_journal_init_inode_cache(void) 3102 { 3103 J_ASSERT(!jbd2_inode_cache); 3104 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); 3105 if (!jbd2_inode_cache) { 3106 pr_emerg("JBD2: failed to create inode cache\n"); 3107 return -ENOMEM; 3108 } 3109 return 0; 3110 } 3111 3112 static int __init jbd2_journal_init_handle_cache(void) 3113 { 3114 J_ASSERT(!jbd2_handle_cache); 3115 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); 3116 if (!jbd2_handle_cache) { 3117 printk(KERN_EMERG "JBD2: failed to create handle cache\n"); 3118 return -ENOMEM; 3119 } 3120 return 0; 3121 } 3122 3123 static void jbd2_journal_destroy_inode_cache(void) 3124 { 3125 kmem_cache_destroy(jbd2_inode_cache); 3126 jbd2_inode_cache = NULL; 3127 } 3128 3129 static void jbd2_journal_destroy_handle_cache(void) 3130 { 3131 kmem_cache_destroy(jbd2_handle_cache); 3132 jbd2_handle_cache = NULL; 3133 } 3134 3135 /* 3136 * Module startup and shutdown 3137 */ 3138 3139 static int __init journal_init_caches(void) 3140 { 3141 int ret; 3142 3143 ret = jbd2_journal_init_revoke_record_cache(); 3144 if (ret == 0) 3145 ret = jbd2_journal_init_revoke_table_cache(); 3146 if (ret == 0) 3147 ret = jbd2_journal_init_journal_head_cache(); 3148 if (ret == 0) 3149 ret = jbd2_journal_init_handle_cache(); 3150 if (ret == 0) 3151 ret = jbd2_journal_init_inode_cache(); 3152 if (ret == 0) 3153 ret = jbd2_journal_init_transaction_cache(); 3154 return ret; 3155 } 3156 3157 static void jbd2_journal_destroy_caches(void) 3158 { 3159 jbd2_journal_destroy_revoke_record_cache(); 3160 jbd2_journal_destroy_revoke_table_cache(); 3161 jbd2_journal_destroy_journal_head_cache(); 3162 jbd2_journal_destroy_handle_cache(); 3163 jbd2_journal_destroy_inode_cache(); 3164 jbd2_journal_destroy_transaction_cache(); 3165 jbd2_journal_destroy_slabs(); 3166 } 3167 3168 static int __init journal_init(void) 3169 { 3170 int ret; 3171 3172 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); 3173 3174 ret = journal_init_caches(); 3175 if (ret == 0) { 3176 jbd2_create_jbd_stats_proc_entry(); 3177 } else { 3178 jbd2_journal_destroy_caches(); 3179 } 3180 return ret; 3181 } 3182 3183 static void __exit journal_exit(void) 3184 { 3185 #ifdef CONFIG_JBD2_DEBUG 3186 int n = atomic_read(&nr_journal_heads); 3187 if (n) 3188 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); 3189 #endif 3190 jbd2_remove_jbd_stats_proc_entry(); 3191 jbd2_journal_destroy_caches(); 3192 } 3193 3194 MODULE_LICENSE("GPL"); 3195 module_init(journal_init); 3196 module_exit(journal_exit); 3197 3198