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