1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * linux/fs/jbd2/journal.c 4 * 5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 6 * 7 * Copyright 1998 Red Hat corp --- All Rights Reserved 8 * 9 * Generic filesystem journal-writing code; part of the ext2fs 10 * journaling system. 11 * 12 * This file manages journals: areas of disk reserved for logging 13 * transactional updates. This includes the kernel journaling thread 14 * which is responsible for scheduling updates to the log. 15 * 16 * We do not actually manage the physical storage of the journal in this 17 * file: that is left to a per-journal policy function, which allows us 18 * to store the journal within a filesystem-specified area for ext2 19 * journaling (ext2 can use a reserved inode for storing the log). 20 */ 21 22 #include <linux/module.h> 23 #include <linux/time.h> 24 #include <linux/fs.h> 25 #include <linux/jbd2.h> 26 #include <linux/errno.h> 27 #include <linux/slab.h> 28 #include <linux/init.h> 29 #include <linux/mm.h> 30 #include <linux/freezer.h> 31 #include <linux/pagemap.h> 32 #include <linux/kthread.h> 33 #include <linux/poison.h> 34 #include <linux/proc_fs.h> 35 #include <linux/seq_file.h> 36 #include <linux/math64.h> 37 #include <linux/hash.h> 38 #include <linux/log2.h> 39 #include <linux/vmalloc.h> 40 #include <linux/backing-dev.h> 41 #include <linux/bitops.h> 42 #include <linux/ratelimit.h> 43 #include <linux/sched/mm.h> 44 45 #define CREATE_TRACE_POINTS 46 #include <trace/events/jbd2.h> 47 48 #include <linux/uaccess.h> 49 #include <asm/page.h> 50 51 #ifdef CONFIG_JBD2_DEBUG 52 static ushort jbd2_journal_enable_debug __read_mostly; 53 54 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644); 55 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2"); 56 #endif 57 58 EXPORT_SYMBOL(jbd2_journal_extend); 59 EXPORT_SYMBOL(jbd2_journal_stop); 60 EXPORT_SYMBOL(jbd2_journal_lock_updates); 61 EXPORT_SYMBOL(jbd2_journal_unlock_updates); 62 EXPORT_SYMBOL(jbd2_journal_get_write_access); 63 EXPORT_SYMBOL(jbd2_journal_get_create_access); 64 EXPORT_SYMBOL(jbd2_journal_get_undo_access); 65 EXPORT_SYMBOL(jbd2_journal_set_triggers); 66 EXPORT_SYMBOL(jbd2_journal_dirty_metadata); 67 EXPORT_SYMBOL(jbd2_journal_forget); 68 EXPORT_SYMBOL(jbd2_journal_flush); 69 EXPORT_SYMBOL(jbd2_journal_revoke); 70 71 EXPORT_SYMBOL(jbd2_journal_init_dev); 72 EXPORT_SYMBOL(jbd2_journal_init_inode); 73 EXPORT_SYMBOL(jbd2_journal_check_used_features); 74 EXPORT_SYMBOL(jbd2_journal_check_available_features); 75 EXPORT_SYMBOL(jbd2_journal_set_features); 76 EXPORT_SYMBOL(jbd2_journal_load); 77 EXPORT_SYMBOL(jbd2_journal_destroy); 78 EXPORT_SYMBOL(jbd2_journal_abort); 79 EXPORT_SYMBOL(jbd2_journal_errno); 80 EXPORT_SYMBOL(jbd2_journal_ack_err); 81 EXPORT_SYMBOL(jbd2_journal_clear_err); 82 EXPORT_SYMBOL(jbd2_log_wait_commit); 83 EXPORT_SYMBOL(jbd2_journal_start_commit); 84 EXPORT_SYMBOL(jbd2_journal_force_commit_nested); 85 EXPORT_SYMBOL(jbd2_journal_wipe); 86 EXPORT_SYMBOL(jbd2_journal_blocks_per_page); 87 EXPORT_SYMBOL(jbd2_journal_invalidate_folio); 88 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); 89 EXPORT_SYMBOL(jbd2_journal_force_commit); 90 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write); 91 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait); 92 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers); 93 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); 94 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); 95 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); 96 EXPORT_SYMBOL(jbd2_inode_cache); 97 98 static int jbd2_journal_create_slab(size_t slab_size); 99 100 #ifdef CONFIG_JBD2_DEBUG 101 void __jbd2_debug(int level, const char *file, const char *func, 102 unsigned int line, const char *fmt, ...) 103 { 104 struct va_format vaf; 105 va_list args; 106 107 if (level > jbd2_journal_enable_debug) 108 return; 109 va_start(args, fmt); 110 vaf.fmt = fmt; 111 vaf.va = &args; 112 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf); 113 va_end(args); 114 } 115 #endif 116 117 /* Checksumming functions */ 118 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) 119 { 120 if (!jbd2_journal_has_csum_v2or3_feature(j)) 121 return 1; 122 123 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; 124 } 125 126 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) 127 { 128 __u32 csum; 129 __be32 old_csum; 130 131 old_csum = sb->s_checksum; 132 sb->s_checksum = 0; 133 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); 134 sb->s_checksum = old_csum; 135 136 return cpu_to_be32(csum); 137 } 138 139 /* 140 * Helper function used to manage commit timeouts 141 */ 142 143 static void commit_timeout(struct timer_list *t) 144 { 145 journal_t *journal = from_timer(journal, t, j_commit_timer); 146 147 wake_up_process(journal->j_task); 148 } 149 150 /* 151 * kjournald2: The main thread function used to manage a logging device 152 * journal. 153 * 154 * This kernel thread is responsible for two things: 155 * 156 * 1) COMMIT: Every so often we need to commit the current state of the 157 * filesystem to disk. The journal thread is responsible for writing 158 * all of the metadata buffers to disk. If a fast commit is ongoing 159 * journal thread waits until it's done and then continues from 160 * there on. 161 * 162 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all 163 * of the data in that part of the log has been rewritten elsewhere on 164 * the disk. Flushing these old buffers to reclaim space in the log is 165 * known as checkpointing, and this thread is responsible for that job. 166 */ 167 168 static int kjournald2(void *arg) 169 { 170 journal_t *journal = arg; 171 transaction_t *transaction; 172 173 /* 174 * Set up an interval timer which can be used to trigger a commit wakeup 175 * after the commit interval expires 176 */ 177 timer_setup(&journal->j_commit_timer, commit_timeout, 0); 178 179 set_freezable(); 180 181 /* Record that the journal thread is running */ 182 journal->j_task = current; 183 wake_up(&journal->j_wait_done_commit); 184 185 /* 186 * Make sure that no allocations from this kernel thread will ever 187 * recurse to the fs layer because we are responsible for the 188 * transaction commit and any fs involvement might get stuck waiting for 189 * the trasn. commit. 190 */ 191 memalloc_nofs_save(); 192 193 /* 194 * And now, wait forever for commit wakeup events. 195 */ 196 write_lock(&journal->j_state_lock); 197 198 loop: 199 if (journal->j_flags & JBD2_UNMOUNT) 200 goto end_loop; 201 202 jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n", 203 journal->j_commit_sequence, journal->j_commit_request); 204 205 if (journal->j_commit_sequence != journal->j_commit_request) { 206 jbd2_debug(1, "OK, requests differ\n"); 207 write_unlock(&journal->j_state_lock); 208 del_timer_sync(&journal->j_commit_timer); 209 jbd2_journal_commit_transaction(journal); 210 write_lock(&journal->j_state_lock); 211 goto loop; 212 } 213 214 wake_up(&journal->j_wait_done_commit); 215 if (freezing(current)) { 216 /* 217 * The simpler the better. Flushing journal isn't a 218 * good idea, because that depends on threads that may 219 * be already stopped. 220 */ 221 jbd2_debug(1, "Now suspending kjournald2\n"); 222 write_unlock(&journal->j_state_lock); 223 try_to_freeze(); 224 write_lock(&journal->j_state_lock); 225 } else { 226 /* 227 * We assume on resume that commits are already there, 228 * so we don't sleep 229 */ 230 DEFINE_WAIT(wait); 231 int should_sleep = 1; 232 233 prepare_to_wait(&journal->j_wait_commit, &wait, 234 TASK_INTERRUPTIBLE); 235 if (journal->j_commit_sequence != journal->j_commit_request) 236 should_sleep = 0; 237 transaction = journal->j_running_transaction; 238 if (transaction && time_after_eq(jiffies, 239 transaction->t_expires)) 240 should_sleep = 0; 241 if (journal->j_flags & JBD2_UNMOUNT) 242 should_sleep = 0; 243 if (should_sleep) { 244 write_unlock(&journal->j_state_lock); 245 schedule(); 246 write_lock(&journal->j_state_lock); 247 } 248 finish_wait(&journal->j_wait_commit, &wait); 249 } 250 251 jbd2_debug(1, "kjournald2 wakes\n"); 252 253 /* 254 * Were we woken up by a commit wakeup event? 255 */ 256 transaction = journal->j_running_transaction; 257 if (transaction && time_after_eq(jiffies, transaction->t_expires)) { 258 journal->j_commit_request = transaction->t_tid; 259 jbd2_debug(1, "woke because of timeout\n"); 260 } 261 goto loop; 262 263 end_loop: 264 del_timer_sync(&journal->j_commit_timer); 265 journal->j_task = NULL; 266 wake_up(&journal->j_wait_done_commit); 267 jbd2_debug(1, "Journal thread exiting.\n"); 268 write_unlock(&journal->j_state_lock); 269 return 0; 270 } 271 272 static int jbd2_journal_start_thread(journal_t *journal) 273 { 274 struct task_struct *t; 275 276 t = kthread_run(kjournald2, journal, "jbd2/%s", 277 journal->j_devname); 278 if (IS_ERR(t)) 279 return PTR_ERR(t); 280 281 wait_event(journal->j_wait_done_commit, journal->j_task != NULL); 282 return 0; 283 } 284 285 static void journal_kill_thread(journal_t *journal) 286 { 287 write_lock(&journal->j_state_lock); 288 journal->j_flags |= JBD2_UNMOUNT; 289 290 while (journal->j_task) { 291 write_unlock(&journal->j_state_lock); 292 wake_up(&journal->j_wait_commit); 293 wait_event(journal->j_wait_done_commit, journal->j_task == NULL); 294 write_lock(&journal->j_state_lock); 295 } 296 write_unlock(&journal->j_state_lock); 297 } 298 299 /* 300 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. 301 * 302 * Writes a metadata buffer to a given disk block. The actual IO is not 303 * performed but a new buffer_head is constructed which labels the data 304 * to be written with the correct destination disk block. 305 * 306 * Any magic-number escaping which needs to be done will cause a 307 * copy-out here. If the buffer happens to start with the 308 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the 309 * magic number is only written to the log for descripter blocks. In 310 * this case, we copy the data and replace the first word with 0, and we 311 * return a result code which indicates that this buffer needs to be 312 * marked as an escaped buffer in the corresponding log descriptor 313 * block. The missing word can then be restored when the block is read 314 * during recovery. 315 * 316 * If the source buffer has already been modified by a new transaction 317 * since we took the last commit snapshot, we use the frozen copy of 318 * that data for IO. If we end up using the existing buffer_head's data 319 * for the write, then we have to make sure nobody modifies it while the 320 * IO is in progress. do_get_write_access() handles this. 321 * 322 * The function returns a pointer to the buffer_head to be used for IO. 323 * 324 * 325 * Return value: 326 * <0: Error 327 * >=0: Finished OK 328 * 329 * On success: 330 * Bit 0 set == escape performed on the data 331 * Bit 1 set == buffer copy-out performed (kfree the data after IO) 332 */ 333 334 int jbd2_journal_write_metadata_buffer(transaction_t *transaction, 335 struct journal_head *jh_in, 336 struct buffer_head **bh_out, 337 sector_t blocknr) 338 { 339 int need_copy_out = 0; 340 int done_copy_out = 0; 341 int do_escape = 0; 342 char *mapped_data; 343 struct buffer_head *new_bh; 344 struct page *new_page; 345 unsigned int new_offset; 346 struct buffer_head *bh_in = jh2bh(jh_in); 347 journal_t *journal = transaction->t_journal; 348 349 /* 350 * The buffer really shouldn't be locked: only the current committing 351 * transaction is allowed to write it, so nobody else is allowed 352 * to do any IO. 353 * 354 * akpm: except if we're journalling data, and write() output is 355 * also part of a shared mapping, and another thread has 356 * decided to launch a writepage() against this buffer. 357 */ 358 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); 359 360 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); 361 362 /* keep subsequent assertions sane */ 363 atomic_set(&new_bh->b_count, 1); 364 365 spin_lock(&jh_in->b_state_lock); 366 repeat: 367 /* 368 * If a new transaction has already done a buffer copy-out, then 369 * we use that version of the data for the commit. 370 */ 371 if (jh_in->b_frozen_data) { 372 done_copy_out = 1; 373 new_page = virt_to_page(jh_in->b_frozen_data); 374 new_offset = offset_in_page(jh_in->b_frozen_data); 375 } else { 376 new_page = jh2bh(jh_in)->b_page; 377 new_offset = offset_in_page(jh2bh(jh_in)->b_data); 378 } 379 380 mapped_data = kmap_atomic(new_page); 381 /* 382 * Fire data frozen trigger if data already wasn't frozen. Do this 383 * before checking for escaping, as the trigger may modify the magic 384 * offset. If a copy-out happens afterwards, it will have the correct 385 * data in the buffer. 386 */ 387 if (!done_copy_out) 388 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, 389 jh_in->b_triggers); 390 391 /* 392 * Check for escaping 393 */ 394 if (*((__be32 *)(mapped_data + new_offset)) == 395 cpu_to_be32(JBD2_MAGIC_NUMBER)) { 396 need_copy_out = 1; 397 do_escape = 1; 398 } 399 kunmap_atomic(mapped_data); 400 401 /* 402 * Do we need to do a data copy? 403 */ 404 if (need_copy_out && !done_copy_out) { 405 char *tmp; 406 407 spin_unlock(&jh_in->b_state_lock); 408 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); 409 if (!tmp) { 410 brelse(new_bh); 411 return -ENOMEM; 412 } 413 spin_lock(&jh_in->b_state_lock); 414 if (jh_in->b_frozen_data) { 415 jbd2_free(tmp, bh_in->b_size); 416 goto repeat; 417 } 418 419 jh_in->b_frozen_data = tmp; 420 mapped_data = kmap_atomic(new_page); 421 memcpy(tmp, mapped_data + new_offset, bh_in->b_size); 422 kunmap_atomic(mapped_data); 423 424 new_page = virt_to_page(tmp); 425 new_offset = offset_in_page(tmp); 426 done_copy_out = 1; 427 428 /* 429 * This isn't strictly necessary, as we're using frozen 430 * data for the escaping, but it keeps consistency with 431 * b_frozen_data usage. 432 */ 433 jh_in->b_frozen_triggers = jh_in->b_triggers; 434 } 435 436 /* 437 * Did we need to do an escaping? Now we've done all the 438 * copying, we can finally do so. 439 */ 440 if (do_escape) { 441 mapped_data = kmap_atomic(new_page); 442 *((unsigned int *)(mapped_data + new_offset)) = 0; 443 kunmap_atomic(mapped_data); 444 } 445 446 set_bh_page(new_bh, new_page, new_offset); 447 new_bh->b_size = bh_in->b_size; 448 new_bh->b_bdev = journal->j_dev; 449 new_bh->b_blocknr = blocknr; 450 new_bh->b_private = bh_in; 451 set_buffer_mapped(new_bh); 452 set_buffer_dirty(new_bh); 453 454 *bh_out = new_bh; 455 456 /* 457 * The to-be-written buffer needs to get moved to the io queue, 458 * and the original buffer whose contents we are shadowing or 459 * copying is moved to the transaction's shadow queue. 460 */ 461 JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); 462 spin_lock(&journal->j_list_lock); 463 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); 464 spin_unlock(&journal->j_list_lock); 465 set_buffer_shadow(bh_in); 466 spin_unlock(&jh_in->b_state_lock); 467 468 return do_escape | (done_copy_out << 1); 469 } 470 471 /* 472 * Allocation code for the journal file. Manage the space left in the 473 * journal, so that we can begin checkpointing when appropriate. 474 */ 475 476 /* 477 * Called with j_state_lock locked for writing. 478 * Returns true if a transaction commit was started. 479 */ 480 static int __jbd2_log_start_commit(journal_t *journal, tid_t target) 481 { 482 /* Return if the txn has already requested to be committed */ 483 if (journal->j_commit_request == target) 484 return 0; 485 486 /* 487 * The only transaction we can possibly wait upon is the 488 * currently running transaction (if it exists). Otherwise, 489 * the target tid must be an old one. 490 */ 491 if (journal->j_running_transaction && 492 journal->j_running_transaction->t_tid == target) { 493 /* 494 * We want a new commit: OK, mark the request and wakeup the 495 * commit thread. We do _not_ do the commit ourselves. 496 */ 497 498 journal->j_commit_request = target; 499 jbd2_debug(1, "JBD2: requesting commit %u/%u\n", 500 journal->j_commit_request, 501 journal->j_commit_sequence); 502 journal->j_running_transaction->t_requested = jiffies; 503 wake_up(&journal->j_wait_commit); 504 return 1; 505 } else if (!tid_geq(journal->j_commit_request, target)) 506 /* This should never happen, but if it does, preserve 507 the evidence before kjournald goes into a loop and 508 increments j_commit_sequence beyond all recognition. */ 509 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", 510 journal->j_commit_request, 511 journal->j_commit_sequence, 512 target, journal->j_running_transaction ? 513 journal->j_running_transaction->t_tid : 0); 514 return 0; 515 } 516 517 int jbd2_log_start_commit(journal_t *journal, tid_t tid) 518 { 519 int ret; 520 521 write_lock(&journal->j_state_lock); 522 ret = __jbd2_log_start_commit(journal, tid); 523 write_unlock(&journal->j_state_lock); 524 return ret; 525 } 526 527 /* 528 * Force and wait any uncommitted transactions. We can only force the running 529 * transaction if we don't have an active handle, otherwise, we will deadlock. 530 * Returns: <0 in case of error, 531 * 0 if nothing to commit, 532 * 1 if transaction was successfully committed. 533 */ 534 static int __jbd2_journal_force_commit(journal_t *journal) 535 { 536 transaction_t *transaction = NULL; 537 tid_t tid; 538 int need_to_start = 0, ret = 0; 539 540 read_lock(&journal->j_state_lock); 541 if (journal->j_running_transaction && !current->journal_info) { 542 transaction = journal->j_running_transaction; 543 if (!tid_geq(journal->j_commit_request, transaction->t_tid)) 544 need_to_start = 1; 545 } else if (journal->j_committing_transaction) 546 transaction = journal->j_committing_transaction; 547 548 if (!transaction) { 549 /* Nothing to commit */ 550 read_unlock(&journal->j_state_lock); 551 return 0; 552 } 553 tid = transaction->t_tid; 554 read_unlock(&journal->j_state_lock); 555 if (need_to_start) 556 jbd2_log_start_commit(journal, tid); 557 ret = jbd2_log_wait_commit(journal, tid); 558 if (!ret) 559 ret = 1; 560 561 return ret; 562 } 563 564 /** 565 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the 566 * calling process is not within transaction. 567 * 568 * @journal: journal to force 569 * Returns true if progress was made. 570 * 571 * This is used for forcing out undo-protected data which contains 572 * bitmaps, when the fs is running out of space. 573 */ 574 int jbd2_journal_force_commit_nested(journal_t *journal) 575 { 576 int ret; 577 578 ret = __jbd2_journal_force_commit(journal); 579 return ret > 0; 580 } 581 582 /** 583 * jbd2_journal_force_commit() - force any uncommitted transactions 584 * @journal: journal to force 585 * 586 * Caller want unconditional commit. We can only force the running transaction 587 * if we don't have an active handle, otherwise, we will deadlock. 588 */ 589 int jbd2_journal_force_commit(journal_t *journal) 590 { 591 int ret; 592 593 J_ASSERT(!current->journal_info); 594 ret = __jbd2_journal_force_commit(journal); 595 if (ret > 0) 596 ret = 0; 597 return ret; 598 } 599 600 /* 601 * Start a commit of the current running transaction (if any). Returns true 602 * if a transaction is going to be committed (or is currently already 603 * committing), and fills its tid in at *ptid 604 */ 605 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) 606 { 607 int ret = 0; 608 609 write_lock(&journal->j_state_lock); 610 if (journal->j_running_transaction) { 611 tid_t tid = journal->j_running_transaction->t_tid; 612 613 __jbd2_log_start_commit(journal, tid); 614 /* There's a running transaction and we've just made sure 615 * it's commit has been scheduled. */ 616 if (ptid) 617 *ptid = tid; 618 ret = 1; 619 } else if (journal->j_committing_transaction) { 620 /* 621 * If commit has been started, then we have to wait for 622 * completion of that transaction. 623 */ 624 if (ptid) 625 *ptid = journal->j_committing_transaction->t_tid; 626 ret = 1; 627 } 628 write_unlock(&journal->j_state_lock); 629 return ret; 630 } 631 632 /* 633 * Return 1 if a given transaction has not yet sent barrier request 634 * connected with a transaction commit. If 0 is returned, transaction 635 * may or may not have sent the barrier. Used to avoid sending barrier 636 * twice in common cases. 637 */ 638 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) 639 { 640 int ret = 0; 641 transaction_t *commit_trans; 642 643 if (!(journal->j_flags & JBD2_BARRIER)) 644 return 0; 645 read_lock(&journal->j_state_lock); 646 /* Transaction already committed? */ 647 if (tid_geq(journal->j_commit_sequence, tid)) 648 goto out; 649 commit_trans = journal->j_committing_transaction; 650 if (!commit_trans || commit_trans->t_tid != tid) { 651 ret = 1; 652 goto out; 653 } 654 /* 655 * Transaction is being committed and we already proceeded to 656 * submitting a flush to fs partition? 657 */ 658 if (journal->j_fs_dev != journal->j_dev) { 659 if (!commit_trans->t_need_data_flush || 660 commit_trans->t_state >= T_COMMIT_DFLUSH) 661 goto out; 662 } else { 663 if (commit_trans->t_state >= T_COMMIT_JFLUSH) 664 goto out; 665 } 666 ret = 1; 667 out: 668 read_unlock(&journal->j_state_lock); 669 return ret; 670 } 671 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); 672 673 /* 674 * Wait for a specified commit to complete. 675 * The caller may not hold the journal lock. 676 */ 677 int jbd2_log_wait_commit(journal_t *journal, tid_t tid) 678 { 679 int err = 0; 680 681 read_lock(&journal->j_state_lock); 682 #ifdef CONFIG_PROVE_LOCKING 683 /* 684 * Some callers make sure transaction is already committing and in that 685 * case we cannot block on open handles anymore. So don't warn in that 686 * case. 687 */ 688 if (tid_gt(tid, journal->j_commit_sequence) && 689 (!journal->j_committing_transaction || 690 journal->j_committing_transaction->t_tid != tid)) { 691 read_unlock(&journal->j_state_lock); 692 jbd2_might_wait_for_commit(journal); 693 read_lock(&journal->j_state_lock); 694 } 695 #endif 696 #ifdef CONFIG_JBD2_DEBUG 697 if (!tid_geq(journal->j_commit_request, tid)) { 698 printk(KERN_ERR 699 "%s: error: j_commit_request=%u, tid=%u\n", 700 __func__, journal->j_commit_request, tid); 701 } 702 #endif 703 while (tid_gt(tid, journal->j_commit_sequence)) { 704 jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n", 705 tid, journal->j_commit_sequence); 706 read_unlock(&journal->j_state_lock); 707 wake_up(&journal->j_wait_commit); 708 wait_event(journal->j_wait_done_commit, 709 !tid_gt(tid, journal->j_commit_sequence)); 710 read_lock(&journal->j_state_lock); 711 } 712 read_unlock(&journal->j_state_lock); 713 714 if (unlikely(is_journal_aborted(journal))) 715 err = -EIO; 716 return err; 717 } 718 719 /* 720 * Start a fast commit. If there's an ongoing fast or full commit wait for 721 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY 722 * if a fast commit is not needed, either because there's an already a commit 723 * going on or this tid has already been committed. Returns -EINVAL if no jbd2 724 * commit has yet been performed. 725 */ 726 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid) 727 { 728 if (unlikely(is_journal_aborted(journal))) 729 return -EIO; 730 /* 731 * Fast commits only allowed if at least one full commit has 732 * been processed. 733 */ 734 if (!journal->j_stats.ts_tid) 735 return -EINVAL; 736 737 write_lock(&journal->j_state_lock); 738 if (tid <= journal->j_commit_sequence) { 739 write_unlock(&journal->j_state_lock); 740 return -EALREADY; 741 } 742 743 if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING || 744 (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) { 745 DEFINE_WAIT(wait); 746 747 prepare_to_wait(&journal->j_fc_wait, &wait, 748 TASK_UNINTERRUPTIBLE); 749 write_unlock(&journal->j_state_lock); 750 schedule(); 751 finish_wait(&journal->j_fc_wait, &wait); 752 return -EALREADY; 753 } 754 journal->j_flags |= JBD2_FAST_COMMIT_ONGOING; 755 write_unlock(&journal->j_state_lock); 756 jbd2_journal_lock_updates(journal); 757 758 return 0; 759 } 760 EXPORT_SYMBOL(jbd2_fc_begin_commit); 761 762 /* 763 * Stop a fast commit. If fallback is set, this function starts commit of 764 * TID tid before any other fast commit can start. 765 */ 766 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback) 767 { 768 jbd2_journal_unlock_updates(journal); 769 if (journal->j_fc_cleanup_callback) 770 journal->j_fc_cleanup_callback(journal, 0, tid); 771 write_lock(&journal->j_state_lock); 772 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING; 773 if (fallback) 774 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING; 775 write_unlock(&journal->j_state_lock); 776 wake_up(&journal->j_fc_wait); 777 if (fallback) 778 return jbd2_complete_transaction(journal, tid); 779 return 0; 780 } 781 782 int jbd2_fc_end_commit(journal_t *journal) 783 { 784 return __jbd2_fc_end_commit(journal, 0, false); 785 } 786 EXPORT_SYMBOL(jbd2_fc_end_commit); 787 788 int jbd2_fc_end_commit_fallback(journal_t *journal) 789 { 790 tid_t tid; 791 792 read_lock(&journal->j_state_lock); 793 tid = journal->j_running_transaction ? 794 journal->j_running_transaction->t_tid : 0; 795 read_unlock(&journal->j_state_lock); 796 return __jbd2_fc_end_commit(journal, tid, true); 797 } 798 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback); 799 800 /* Return 1 when transaction with given tid has already committed. */ 801 int jbd2_transaction_committed(journal_t *journal, tid_t tid) 802 { 803 int ret = 1; 804 805 read_lock(&journal->j_state_lock); 806 if (journal->j_running_transaction && 807 journal->j_running_transaction->t_tid == tid) 808 ret = 0; 809 if (journal->j_committing_transaction && 810 journal->j_committing_transaction->t_tid == tid) 811 ret = 0; 812 read_unlock(&journal->j_state_lock); 813 return ret; 814 } 815 EXPORT_SYMBOL(jbd2_transaction_committed); 816 817 /* 818 * When this function returns the transaction corresponding to tid 819 * will be completed. If the transaction has currently running, start 820 * committing that transaction before waiting for it to complete. If 821 * the transaction id is stale, it is by definition already completed, 822 * so just return SUCCESS. 823 */ 824 int jbd2_complete_transaction(journal_t *journal, tid_t tid) 825 { 826 int need_to_wait = 1; 827 828 read_lock(&journal->j_state_lock); 829 if (journal->j_running_transaction && 830 journal->j_running_transaction->t_tid == tid) { 831 if (journal->j_commit_request != tid) { 832 /* transaction not yet started, so request it */ 833 read_unlock(&journal->j_state_lock); 834 jbd2_log_start_commit(journal, tid); 835 goto wait_commit; 836 } 837 } else if (!(journal->j_committing_transaction && 838 journal->j_committing_transaction->t_tid == tid)) 839 need_to_wait = 0; 840 read_unlock(&journal->j_state_lock); 841 if (!need_to_wait) 842 return 0; 843 wait_commit: 844 return jbd2_log_wait_commit(journal, tid); 845 } 846 EXPORT_SYMBOL(jbd2_complete_transaction); 847 848 /* 849 * Log buffer allocation routines: 850 */ 851 852 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) 853 { 854 unsigned long blocknr; 855 856 write_lock(&journal->j_state_lock); 857 J_ASSERT(journal->j_free > 1); 858 859 blocknr = journal->j_head; 860 journal->j_head++; 861 journal->j_free--; 862 if (journal->j_head == journal->j_last) 863 journal->j_head = journal->j_first; 864 write_unlock(&journal->j_state_lock); 865 return jbd2_journal_bmap(journal, blocknr, retp); 866 } 867 868 /* Map one fast commit buffer for use by the file system */ 869 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out) 870 { 871 unsigned long long pblock; 872 unsigned long blocknr; 873 int ret = 0; 874 struct buffer_head *bh; 875 int fc_off; 876 877 *bh_out = NULL; 878 879 if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) { 880 fc_off = journal->j_fc_off; 881 blocknr = journal->j_fc_first + fc_off; 882 journal->j_fc_off++; 883 } else { 884 ret = -EINVAL; 885 } 886 887 if (ret) 888 return ret; 889 890 ret = jbd2_journal_bmap(journal, blocknr, &pblock); 891 if (ret) 892 return ret; 893 894 bh = __getblk(journal->j_dev, pblock, journal->j_blocksize); 895 if (!bh) 896 return -ENOMEM; 897 898 899 journal->j_fc_wbuf[fc_off] = bh; 900 901 *bh_out = bh; 902 903 return 0; 904 } 905 EXPORT_SYMBOL(jbd2_fc_get_buf); 906 907 /* 908 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf 909 * for completion. 910 */ 911 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks) 912 { 913 struct buffer_head *bh; 914 int i, j_fc_off; 915 916 j_fc_off = journal->j_fc_off; 917 918 /* 919 * Wait in reverse order to minimize chances of us being woken up before 920 * all IOs have completed 921 */ 922 for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) { 923 bh = journal->j_fc_wbuf[i]; 924 wait_on_buffer(bh); 925 /* 926 * Update j_fc_off so jbd2_fc_release_bufs can release remain 927 * buffer head. 928 */ 929 if (unlikely(!buffer_uptodate(bh))) { 930 journal->j_fc_off = i + 1; 931 return -EIO; 932 } 933 put_bh(bh); 934 journal->j_fc_wbuf[i] = NULL; 935 } 936 937 return 0; 938 } 939 EXPORT_SYMBOL(jbd2_fc_wait_bufs); 940 941 int jbd2_fc_release_bufs(journal_t *journal) 942 { 943 struct buffer_head *bh; 944 int i, j_fc_off; 945 946 j_fc_off = journal->j_fc_off; 947 948 for (i = j_fc_off - 1; i >= 0; i--) { 949 bh = journal->j_fc_wbuf[i]; 950 if (!bh) 951 break; 952 put_bh(bh); 953 journal->j_fc_wbuf[i] = NULL; 954 } 955 956 return 0; 957 } 958 EXPORT_SYMBOL(jbd2_fc_release_bufs); 959 960 /* 961 * Conversion of logical to physical block numbers for the journal 962 * 963 * On external journals the journal blocks are identity-mapped, so 964 * this is a no-op. If needed, we can use j_blk_offset - everything is 965 * ready. 966 */ 967 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, 968 unsigned long long *retp) 969 { 970 int err = 0; 971 unsigned long long ret; 972 sector_t block = blocknr; 973 974 if (journal->j_bmap) { 975 err = journal->j_bmap(journal, &block); 976 if (err == 0) 977 *retp = block; 978 } else if (journal->j_inode) { 979 ret = bmap(journal->j_inode, &block); 980 981 if (ret || !block) { 982 printk(KERN_ALERT "%s: journal block not found " 983 "at offset %lu on %s\n", 984 __func__, blocknr, journal->j_devname); 985 err = -EIO; 986 jbd2_journal_abort(journal, err); 987 } else { 988 *retp = block; 989 } 990 991 } else { 992 *retp = blocknr; /* +journal->j_blk_offset */ 993 } 994 return err; 995 } 996 997 /* 998 * We play buffer_head aliasing tricks to write data/metadata blocks to 999 * the journal without copying their contents, but for journal 1000 * descriptor blocks we do need to generate bona fide buffers. 1001 * 1002 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying 1003 * the buffer's contents they really should run flush_dcache_page(bh->b_page). 1004 * But we don't bother doing that, so there will be coherency problems with 1005 * mmaps of blockdevs which hold live JBD-controlled filesystems. 1006 */ 1007 struct buffer_head * 1008 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type) 1009 { 1010 journal_t *journal = transaction->t_journal; 1011 struct buffer_head *bh; 1012 unsigned long long blocknr; 1013 journal_header_t *header; 1014 int err; 1015 1016 err = jbd2_journal_next_log_block(journal, &blocknr); 1017 1018 if (err) 1019 return NULL; 1020 1021 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); 1022 if (!bh) 1023 return NULL; 1024 atomic_dec(&transaction->t_outstanding_credits); 1025 lock_buffer(bh); 1026 memset(bh->b_data, 0, journal->j_blocksize); 1027 header = (journal_header_t *)bh->b_data; 1028 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); 1029 header->h_blocktype = cpu_to_be32(type); 1030 header->h_sequence = cpu_to_be32(transaction->t_tid); 1031 set_buffer_uptodate(bh); 1032 unlock_buffer(bh); 1033 BUFFER_TRACE(bh, "return this buffer"); 1034 return bh; 1035 } 1036 1037 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh) 1038 { 1039 struct jbd2_journal_block_tail *tail; 1040 __u32 csum; 1041 1042 if (!jbd2_journal_has_csum_v2or3(j)) 1043 return; 1044 1045 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - 1046 sizeof(struct jbd2_journal_block_tail)); 1047 tail->t_checksum = 0; 1048 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); 1049 tail->t_checksum = cpu_to_be32(csum); 1050 } 1051 1052 /* 1053 * Return tid of the oldest transaction in the journal and block in the journal 1054 * where the transaction starts. 1055 * 1056 * If the journal is now empty, return which will be the next transaction ID 1057 * we will write and where will that transaction start. 1058 * 1059 * The return value is 0 if journal tail cannot be pushed any further, 1 if 1060 * it can. 1061 */ 1062 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, 1063 unsigned long *block) 1064 { 1065 transaction_t *transaction; 1066 int ret; 1067 1068 read_lock(&journal->j_state_lock); 1069 spin_lock(&journal->j_list_lock); 1070 transaction = journal->j_checkpoint_transactions; 1071 if (transaction) { 1072 *tid = transaction->t_tid; 1073 *block = transaction->t_log_start; 1074 } else if ((transaction = journal->j_committing_transaction) != NULL) { 1075 *tid = transaction->t_tid; 1076 *block = transaction->t_log_start; 1077 } else if ((transaction = journal->j_running_transaction) != NULL) { 1078 *tid = transaction->t_tid; 1079 *block = journal->j_head; 1080 } else { 1081 *tid = journal->j_transaction_sequence; 1082 *block = journal->j_head; 1083 } 1084 ret = tid_gt(*tid, journal->j_tail_sequence); 1085 spin_unlock(&journal->j_list_lock); 1086 read_unlock(&journal->j_state_lock); 1087 1088 return ret; 1089 } 1090 1091 /* 1092 * Update information in journal structure and in on disk journal superblock 1093 * about log tail. This function does not check whether information passed in 1094 * really pushes log tail further. It's responsibility of the caller to make 1095 * sure provided log tail information is valid (e.g. by holding 1096 * j_checkpoint_mutex all the time between computing log tail and calling this 1097 * function as is the case with jbd2_cleanup_journal_tail()). 1098 * 1099 * Requires j_checkpoint_mutex 1100 */ 1101 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1102 { 1103 unsigned long freed; 1104 int ret; 1105 1106 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1107 1108 /* 1109 * We cannot afford for write to remain in drive's caches since as 1110 * soon as we update j_tail, next transaction can start reusing journal 1111 * space and if we lose sb update during power failure we'd replay 1112 * old transaction with possibly newly overwritten data. 1113 */ 1114 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, 1115 REQ_SYNC | REQ_FUA); 1116 if (ret) 1117 goto out; 1118 1119 write_lock(&journal->j_state_lock); 1120 freed = block - journal->j_tail; 1121 if (block < journal->j_tail) 1122 freed += journal->j_last - journal->j_first; 1123 1124 trace_jbd2_update_log_tail(journal, tid, block, freed); 1125 jbd2_debug(1, 1126 "Cleaning journal tail from %u to %u (offset %lu), " 1127 "freeing %lu\n", 1128 journal->j_tail_sequence, tid, block, freed); 1129 1130 journal->j_free += freed; 1131 journal->j_tail_sequence = tid; 1132 journal->j_tail = block; 1133 write_unlock(&journal->j_state_lock); 1134 1135 out: 1136 return ret; 1137 } 1138 1139 /* 1140 * This is a variation of __jbd2_update_log_tail which checks for validity of 1141 * provided log tail and locks j_checkpoint_mutex. So it is safe against races 1142 * with other threads updating log tail. 1143 */ 1144 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1145 { 1146 mutex_lock_io(&journal->j_checkpoint_mutex); 1147 if (tid_gt(tid, journal->j_tail_sequence)) 1148 __jbd2_update_log_tail(journal, tid, block); 1149 mutex_unlock(&journal->j_checkpoint_mutex); 1150 } 1151 1152 struct jbd2_stats_proc_session { 1153 journal_t *journal; 1154 struct transaction_stats_s *stats; 1155 int start; 1156 int max; 1157 }; 1158 1159 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) 1160 { 1161 return *pos ? NULL : SEQ_START_TOKEN; 1162 } 1163 1164 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) 1165 { 1166 (*pos)++; 1167 return NULL; 1168 } 1169 1170 static int jbd2_seq_info_show(struct seq_file *seq, void *v) 1171 { 1172 struct jbd2_stats_proc_session *s = seq->private; 1173 1174 if (v != SEQ_START_TOKEN) 1175 return 0; 1176 seq_printf(seq, "%lu transactions (%lu requested), " 1177 "each up to %u blocks\n", 1178 s->stats->ts_tid, s->stats->ts_requested, 1179 s->journal->j_max_transaction_buffers); 1180 if (s->stats->ts_tid == 0) 1181 return 0; 1182 seq_printf(seq, "average: \n %ums waiting for transaction\n", 1183 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); 1184 seq_printf(seq, " %ums request delay\n", 1185 (s->stats->ts_requested == 0) ? 0 : 1186 jiffies_to_msecs(s->stats->run.rs_request_delay / 1187 s->stats->ts_requested)); 1188 seq_printf(seq, " %ums running transaction\n", 1189 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); 1190 seq_printf(seq, " %ums transaction was being locked\n", 1191 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); 1192 seq_printf(seq, " %ums flushing data (in ordered mode)\n", 1193 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); 1194 seq_printf(seq, " %ums logging transaction\n", 1195 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); 1196 seq_printf(seq, " %lluus average transaction commit time\n", 1197 div_u64(s->journal->j_average_commit_time, 1000)); 1198 seq_printf(seq, " %lu handles per transaction\n", 1199 s->stats->run.rs_handle_count / s->stats->ts_tid); 1200 seq_printf(seq, " %lu blocks per transaction\n", 1201 s->stats->run.rs_blocks / s->stats->ts_tid); 1202 seq_printf(seq, " %lu logged blocks per transaction\n", 1203 s->stats->run.rs_blocks_logged / s->stats->ts_tid); 1204 return 0; 1205 } 1206 1207 static void jbd2_seq_info_stop(struct seq_file *seq, void *v) 1208 { 1209 } 1210 1211 static const struct seq_operations jbd2_seq_info_ops = { 1212 .start = jbd2_seq_info_start, 1213 .next = jbd2_seq_info_next, 1214 .stop = jbd2_seq_info_stop, 1215 .show = jbd2_seq_info_show, 1216 }; 1217 1218 static int jbd2_seq_info_open(struct inode *inode, struct file *file) 1219 { 1220 journal_t *journal = pde_data(inode); 1221 struct jbd2_stats_proc_session *s; 1222 int rc, size; 1223 1224 s = kmalloc(sizeof(*s), GFP_KERNEL); 1225 if (s == NULL) 1226 return -ENOMEM; 1227 size = sizeof(struct transaction_stats_s); 1228 s->stats = kmalloc(size, GFP_KERNEL); 1229 if (s->stats == NULL) { 1230 kfree(s); 1231 return -ENOMEM; 1232 } 1233 spin_lock(&journal->j_history_lock); 1234 memcpy(s->stats, &journal->j_stats, size); 1235 s->journal = journal; 1236 spin_unlock(&journal->j_history_lock); 1237 1238 rc = seq_open(file, &jbd2_seq_info_ops); 1239 if (rc == 0) { 1240 struct seq_file *m = file->private_data; 1241 m->private = s; 1242 } else { 1243 kfree(s->stats); 1244 kfree(s); 1245 } 1246 return rc; 1247 1248 } 1249 1250 static int jbd2_seq_info_release(struct inode *inode, struct file *file) 1251 { 1252 struct seq_file *seq = file->private_data; 1253 struct jbd2_stats_proc_session *s = seq->private; 1254 kfree(s->stats); 1255 kfree(s); 1256 return seq_release(inode, file); 1257 } 1258 1259 static const struct proc_ops jbd2_info_proc_ops = { 1260 .proc_open = jbd2_seq_info_open, 1261 .proc_read = seq_read, 1262 .proc_lseek = seq_lseek, 1263 .proc_release = jbd2_seq_info_release, 1264 }; 1265 1266 static struct proc_dir_entry *proc_jbd2_stats; 1267 1268 static void jbd2_stats_proc_init(journal_t *journal) 1269 { 1270 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); 1271 if (journal->j_proc_entry) { 1272 proc_create_data("info", S_IRUGO, journal->j_proc_entry, 1273 &jbd2_info_proc_ops, journal); 1274 } 1275 } 1276 1277 static void jbd2_stats_proc_exit(journal_t *journal) 1278 { 1279 remove_proc_entry("info", journal->j_proc_entry); 1280 remove_proc_entry(journal->j_devname, proc_jbd2_stats); 1281 } 1282 1283 /* Minimum size of descriptor tag */ 1284 static int jbd2_min_tag_size(void) 1285 { 1286 /* 1287 * Tag with 32-bit block numbers does not use last four bytes of the 1288 * structure 1289 */ 1290 return sizeof(journal_block_tag_t) - 4; 1291 } 1292 1293 /** 1294 * jbd2_journal_shrink_scan() 1295 * @shrink: shrinker to work on 1296 * @sc: reclaim request to process 1297 * 1298 * Scan the checkpointed buffer on the checkpoint list and release the 1299 * journal_head. 1300 */ 1301 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink, 1302 struct shrink_control *sc) 1303 { 1304 journal_t *journal = container_of(shrink, journal_t, j_shrinker); 1305 unsigned long nr_to_scan = sc->nr_to_scan; 1306 unsigned long nr_shrunk; 1307 unsigned long count; 1308 1309 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1310 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count); 1311 1312 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan); 1313 1314 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1315 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count); 1316 1317 return nr_shrunk; 1318 } 1319 1320 /** 1321 * jbd2_journal_shrink_count() 1322 * @shrink: shrinker to work on 1323 * @sc: reclaim request to process 1324 * 1325 * Count the number of checkpoint buffers on the checkpoint list. 1326 */ 1327 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink, 1328 struct shrink_control *sc) 1329 { 1330 journal_t *journal = container_of(shrink, journal_t, j_shrinker); 1331 unsigned long count; 1332 1333 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1334 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count); 1335 1336 return count; 1337 } 1338 1339 /* 1340 * Management for journal control blocks: functions to create and 1341 * destroy journal_t structures, and to initialise and read existing 1342 * journal blocks from disk. */ 1343 1344 /* First: create and setup a journal_t object in memory. We initialise 1345 * very few fields yet: that has to wait until we have created the 1346 * journal structures from from scratch, or loaded them from disk. */ 1347 1348 static journal_t *journal_init_common(struct block_device *bdev, 1349 struct block_device *fs_dev, 1350 unsigned long long start, int len, int blocksize) 1351 { 1352 static struct lock_class_key jbd2_trans_commit_key; 1353 journal_t *journal; 1354 int err; 1355 struct buffer_head *bh; 1356 int n; 1357 1358 journal = kzalloc(sizeof(*journal), GFP_KERNEL); 1359 if (!journal) 1360 return NULL; 1361 1362 init_waitqueue_head(&journal->j_wait_transaction_locked); 1363 init_waitqueue_head(&journal->j_wait_done_commit); 1364 init_waitqueue_head(&journal->j_wait_commit); 1365 init_waitqueue_head(&journal->j_wait_updates); 1366 init_waitqueue_head(&journal->j_wait_reserved); 1367 init_waitqueue_head(&journal->j_fc_wait); 1368 mutex_init(&journal->j_abort_mutex); 1369 mutex_init(&journal->j_barrier); 1370 mutex_init(&journal->j_checkpoint_mutex); 1371 spin_lock_init(&journal->j_revoke_lock); 1372 spin_lock_init(&journal->j_list_lock); 1373 rwlock_init(&journal->j_state_lock); 1374 1375 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); 1376 journal->j_min_batch_time = 0; 1377 journal->j_max_batch_time = 15000; /* 15ms */ 1378 atomic_set(&journal->j_reserved_credits, 0); 1379 1380 /* The journal is marked for error until we succeed with recovery! */ 1381 journal->j_flags = JBD2_ABORT; 1382 1383 /* Set up a default-sized revoke table for the new mount. */ 1384 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); 1385 if (err) 1386 goto err_cleanup; 1387 1388 spin_lock_init(&journal->j_history_lock); 1389 1390 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle", 1391 &jbd2_trans_commit_key, 0); 1392 1393 /* journal descriptor can store up to n blocks -bzzz */ 1394 journal->j_blocksize = blocksize; 1395 journal->j_dev = bdev; 1396 journal->j_fs_dev = fs_dev; 1397 journal->j_blk_offset = start; 1398 journal->j_total_len = len; 1399 /* We need enough buffers to write out full descriptor block. */ 1400 n = journal->j_blocksize / jbd2_min_tag_size(); 1401 journal->j_wbufsize = n; 1402 journal->j_fc_wbuf = NULL; 1403 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *), 1404 GFP_KERNEL); 1405 if (!journal->j_wbuf) 1406 goto err_cleanup; 1407 1408 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize); 1409 if (!bh) { 1410 pr_err("%s: Cannot get buffer for journal superblock\n", 1411 __func__); 1412 goto err_cleanup; 1413 } 1414 journal->j_sb_buffer = bh; 1415 journal->j_superblock = (journal_superblock_t *)bh->b_data; 1416 1417 journal->j_shrink_transaction = NULL; 1418 journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan; 1419 journal->j_shrinker.count_objects = jbd2_journal_shrink_count; 1420 journal->j_shrinker.seeks = DEFAULT_SEEKS; 1421 journal->j_shrinker.batch = journal->j_max_transaction_buffers; 1422 1423 if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL)) 1424 goto err_cleanup; 1425 1426 if (register_shrinker(&journal->j_shrinker, "jbd2-journal:(%u:%u)", 1427 MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev))) { 1428 percpu_counter_destroy(&journal->j_checkpoint_jh_count); 1429 goto err_cleanup; 1430 } 1431 return journal; 1432 1433 err_cleanup: 1434 brelse(journal->j_sb_buffer); 1435 kfree(journal->j_wbuf); 1436 jbd2_journal_destroy_revoke(journal); 1437 kfree(journal); 1438 return NULL; 1439 } 1440 1441 /* jbd2_journal_init_dev and jbd2_journal_init_inode: 1442 * 1443 * Create a journal structure assigned some fixed set of disk blocks to 1444 * the journal. We don't actually touch those disk blocks yet, but we 1445 * need to set up all of the mapping information to tell the journaling 1446 * system where the journal blocks are. 1447 * 1448 */ 1449 1450 /** 1451 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure 1452 * @bdev: Block device on which to create the journal 1453 * @fs_dev: Device which hold journalled filesystem for this journal. 1454 * @start: Block nr Start of journal. 1455 * @len: Length of the journal in blocks. 1456 * @blocksize: blocksize of journalling device 1457 * 1458 * Returns: a newly created journal_t * 1459 * 1460 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous 1461 * range of blocks on an arbitrary block device. 1462 * 1463 */ 1464 journal_t *jbd2_journal_init_dev(struct block_device *bdev, 1465 struct block_device *fs_dev, 1466 unsigned long long start, int len, int blocksize) 1467 { 1468 journal_t *journal; 1469 1470 journal = journal_init_common(bdev, fs_dev, start, len, blocksize); 1471 if (!journal) 1472 return NULL; 1473 1474 snprintf(journal->j_devname, sizeof(journal->j_devname), 1475 "%pg", journal->j_dev); 1476 strreplace(journal->j_devname, '/', '!'); 1477 jbd2_stats_proc_init(journal); 1478 1479 return journal; 1480 } 1481 1482 /** 1483 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. 1484 * @inode: An inode to create the journal in 1485 * 1486 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as 1487 * the journal. The inode must exist already, must support bmap() and 1488 * must have all data blocks preallocated. 1489 */ 1490 journal_t *jbd2_journal_init_inode(struct inode *inode) 1491 { 1492 journal_t *journal; 1493 sector_t blocknr; 1494 int err = 0; 1495 1496 blocknr = 0; 1497 err = bmap(inode, &blocknr); 1498 1499 if (err || !blocknr) { 1500 pr_err("%s: Cannot locate journal superblock\n", 1501 __func__); 1502 return NULL; 1503 } 1504 1505 jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n", 1506 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size, 1507 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); 1508 1509 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev, 1510 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits, 1511 inode->i_sb->s_blocksize); 1512 if (!journal) 1513 return NULL; 1514 1515 journal->j_inode = inode; 1516 snprintf(journal->j_devname, sizeof(journal->j_devname), 1517 "%pg-%lu", journal->j_dev, journal->j_inode->i_ino); 1518 strreplace(journal->j_devname, '/', '!'); 1519 jbd2_stats_proc_init(journal); 1520 1521 return journal; 1522 } 1523 1524 /* 1525 * If the journal init or create aborts, we need to mark the journal 1526 * superblock as being NULL to prevent the journal destroy from writing 1527 * back a bogus superblock. 1528 */ 1529 static void journal_fail_superblock(journal_t *journal) 1530 { 1531 struct buffer_head *bh = journal->j_sb_buffer; 1532 brelse(bh); 1533 journal->j_sb_buffer = NULL; 1534 } 1535 1536 /* 1537 * Given a journal_t structure, initialise the various fields for 1538 * startup of a new journaling session. We use this both when creating 1539 * a journal, and after recovering an old journal to reset it for 1540 * subsequent use. 1541 */ 1542 1543 static int journal_reset(journal_t *journal) 1544 { 1545 journal_superblock_t *sb = journal->j_superblock; 1546 unsigned long long first, last; 1547 1548 first = be32_to_cpu(sb->s_first); 1549 last = be32_to_cpu(sb->s_maxlen); 1550 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { 1551 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", 1552 first, last); 1553 journal_fail_superblock(journal); 1554 return -EINVAL; 1555 } 1556 1557 journal->j_first = first; 1558 journal->j_last = last; 1559 1560 journal->j_head = journal->j_first; 1561 journal->j_tail = journal->j_first; 1562 journal->j_free = journal->j_last - journal->j_first; 1563 1564 journal->j_tail_sequence = journal->j_transaction_sequence; 1565 journal->j_commit_sequence = journal->j_transaction_sequence - 1; 1566 journal->j_commit_request = journal->j_commit_sequence; 1567 1568 journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal); 1569 1570 /* 1571 * Now that journal recovery is done, turn fast commits off here. This 1572 * way, if fast commit was enabled before the crash but if now FS has 1573 * disabled it, we don't enable fast commits. 1574 */ 1575 jbd2_clear_feature_fast_commit(journal); 1576 1577 /* 1578 * As a special case, if the on-disk copy is already marked as needing 1579 * no recovery (s_start == 0), then we can safely defer the superblock 1580 * update until the next commit by setting JBD2_FLUSHED. This avoids 1581 * attempting a write to a potential-readonly device. 1582 */ 1583 if (sb->s_start == 0) { 1584 jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb " 1585 "(start %ld, seq %u, errno %d)\n", 1586 journal->j_tail, journal->j_tail_sequence, 1587 journal->j_errno); 1588 journal->j_flags |= JBD2_FLUSHED; 1589 } else { 1590 /* Lock here to make assertions happy... */ 1591 mutex_lock_io(&journal->j_checkpoint_mutex); 1592 /* 1593 * Update log tail information. We use REQ_FUA since new 1594 * transaction will start reusing journal space and so we 1595 * must make sure information about current log tail is on 1596 * disk before that. 1597 */ 1598 jbd2_journal_update_sb_log_tail(journal, 1599 journal->j_tail_sequence, 1600 journal->j_tail, 1601 REQ_SYNC | REQ_FUA); 1602 mutex_unlock(&journal->j_checkpoint_mutex); 1603 } 1604 return jbd2_journal_start_thread(journal); 1605 } 1606 1607 /* 1608 * This function expects that the caller will have locked the journal 1609 * buffer head, and will return with it unlocked 1610 */ 1611 static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags) 1612 { 1613 struct buffer_head *bh = journal->j_sb_buffer; 1614 journal_superblock_t *sb = journal->j_superblock; 1615 int ret = 0; 1616 1617 /* Buffer got discarded which means block device got invalidated */ 1618 if (!buffer_mapped(bh)) { 1619 unlock_buffer(bh); 1620 return -EIO; 1621 } 1622 1623 trace_jbd2_write_superblock(journal, write_flags); 1624 if (!(journal->j_flags & JBD2_BARRIER)) 1625 write_flags &= ~(REQ_FUA | REQ_PREFLUSH); 1626 if (buffer_write_io_error(bh)) { 1627 /* 1628 * Oh, dear. A previous attempt to write the journal 1629 * superblock failed. This could happen because the 1630 * USB device was yanked out. Or it could happen to 1631 * be a transient write error and maybe the block will 1632 * be remapped. Nothing we can do but to retry the 1633 * write and hope for the best. 1634 */ 1635 printk(KERN_ERR "JBD2: previous I/O error detected " 1636 "for journal superblock update for %s.\n", 1637 journal->j_devname); 1638 clear_buffer_write_io_error(bh); 1639 set_buffer_uptodate(bh); 1640 } 1641 if (jbd2_journal_has_csum_v2or3(journal)) 1642 sb->s_checksum = jbd2_superblock_csum(journal, sb); 1643 get_bh(bh); 1644 bh->b_end_io = end_buffer_write_sync; 1645 submit_bh(REQ_OP_WRITE | write_flags, bh); 1646 wait_on_buffer(bh); 1647 if (buffer_write_io_error(bh)) { 1648 clear_buffer_write_io_error(bh); 1649 set_buffer_uptodate(bh); 1650 ret = -EIO; 1651 } 1652 if (ret) { 1653 printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n", 1654 journal->j_devname); 1655 if (!is_journal_aborted(journal)) 1656 jbd2_journal_abort(journal, ret); 1657 } 1658 1659 return ret; 1660 } 1661 1662 /** 1663 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. 1664 * @journal: The journal to update. 1665 * @tail_tid: TID of the new transaction at the tail of the log 1666 * @tail_block: The first block of the transaction at the tail of the log 1667 * @write_flags: Flags for the journal sb write operation 1668 * 1669 * Update a journal's superblock information about log tail and write it to 1670 * disk, waiting for the IO to complete. 1671 */ 1672 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, 1673 unsigned long tail_block, 1674 blk_opf_t write_flags) 1675 { 1676 journal_superblock_t *sb = journal->j_superblock; 1677 int ret; 1678 1679 if (is_journal_aborted(journal)) 1680 return -EIO; 1681 if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) { 1682 jbd2_journal_abort(journal, -EIO); 1683 return -EIO; 1684 } 1685 1686 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1687 jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", 1688 tail_block, tail_tid); 1689 1690 lock_buffer(journal->j_sb_buffer); 1691 sb->s_sequence = cpu_to_be32(tail_tid); 1692 sb->s_start = cpu_to_be32(tail_block); 1693 1694 ret = jbd2_write_superblock(journal, write_flags); 1695 if (ret) 1696 goto out; 1697 1698 /* Log is no longer empty */ 1699 write_lock(&journal->j_state_lock); 1700 WARN_ON(!sb->s_sequence); 1701 journal->j_flags &= ~JBD2_FLUSHED; 1702 write_unlock(&journal->j_state_lock); 1703 1704 out: 1705 return ret; 1706 } 1707 1708 /** 1709 * jbd2_mark_journal_empty() - Mark on disk journal as empty. 1710 * @journal: The journal to update. 1711 * @write_flags: Flags for the journal sb write operation 1712 * 1713 * Update a journal's dynamic superblock fields to show that journal is empty. 1714 * Write updated superblock to disk waiting for IO to complete. 1715 */ 1716 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags) 1717 { 1718 journal_superblock_t *sb = journal->j_superblock; 1719 bool had_fast_commit = false; 1720 1721 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1722 lock_buffer(journal->j_sb_buffer); 1723 if (sb->s_start == 0) { /* Is it already empty? */ 1724 unlock_buffer(journal->j_sb_buffer); 1725 return; 1726 } 1727 1728 jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n", 1729 journal->j_tail_sequence); 1730 1731 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); 1732 sb->s_start = cpu_to_be32(0); 1733 if (jbd2_has_feature_fast_commit(journal)) { 1734 /* 1735 * When journal is clean, no need to commit fast commit flag and 1736 * make file system incompatible with older kernels. 1737 */ 1738 jbd2_clear_feature_fast_commit(journal); 1739 had_fast_commit = true; 1740 } 1741 1742 jbd2_write_superblock(journal, write_flags); 1743 1744 if (had_fast_commit) 1745 jbd2_set_feature_fast_commit(journal); 1746 1747 /* Log is no longer empty */ 1748 write_lock(&journal->j_state_lock); 1749 journal->j_flags |= JBD2_FLUSHED; 1750 write_unlock(&journal->j_state_lock); 1751 } 1752 1753 /** 1754 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock) 1755 * @journal: The journal to erase. 1756 * @flags: A discard/zeroout request is sent for each physically contigous 1757 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or 1758 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation 1759 * to perform. 1760 * 1761 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes 1762 * will be explicitly written if no hardware offload is available, see 1763 * blkdev_issue_zeroout for more details. 1764 */ 1765 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags) 1766 { 1767 int err = 0; 1768 unsigned long block, log_offset; /* logical */ 1769 unsigned long long phys_block, block_start, block_stop; /* physical */ 1770 loff_t byte_start, byte_stop, byte_count; 1771 1772 /* flags must be set to either discard or zeroout */ 1773 if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags || 1774 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && 1775 (flags & JBD2_JOURNAL_FLUSH_ZEROOUT))) 1776 return -EINVAL; 1777 1778 if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && 1779 !bdev_max_discard_sectors(journal->j_dev)) 1780 return -EOPNOTSUPP; 1781 1782 /* 1783 * lookup block mapping and issue discard/zeroout for each 1784 * contiguous region 1785 */ 1786 log_offset = be32_to_cpu(journal->j_superblock->s_first); 1787 block_start = ~0ULL; 1788 for (block = log_offset; block < journal->j_total_len; block++) { 1789 err = jbd2_journal_bmap(journal, block, &phys_block); 1790 if (err) { 1791 pr_err("JBD2: bad block at offset %lu", block); 1792 return err; 1793 } 1794 1795 if (block_start == ~0ULL) { 1796 block_start = phys_block; 1797 block_stop = block_start - 1; 1798 } 1799 1800 /* 1801 * last block not contiguous with current block, 1802 * process last contiguous region and return to this block on 1803 * next loop 1804 */ 1805 if (phys_block != block_stop + 1) { 1806 block--; 1807 } else { 1808 block_stop++; 1809 /* 1810 * if this isn't the last block of journal, 1811 * no need to process now because next block may also 1812 * be part of this contiguous region 1813 */ 1814 if (block != journal->j_total_len - 1) 1815 continue; 1816 } 1817 1818 /* 1819 * end of contiguous region or this is last block of journal, 1820 * take care of the region 1821 */ 1822 byte_start = block_start * journal->j_blocksize; 1823 byte_stop = block_stop * journal->j_blocksize; 1824 byte_count = (block_stop - block_start + 1) * 1825 journal->j_blocksize; 1826 1827 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping, 1828 byte_start, byte_stop); 1829 1830 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) { 1831 err = blkdev_issue_discard(journal->j_dev, 1832 byte_start >> SECTOR_SHIFT, 1833 byte_count >> SECTOR_SHIFT, 1834 GFP_NOFS); 1835 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) { 1836 err = blkdev_issue_zeroout(journal->j_dev, 1837 byte_start >> SECTOR_SHIFT, 1838 byte_count >> SECTOR_SHIFT, 1839 GFP_NOFS, 0); 1840 } 1841 1842 if (unlikely(err != 0)) { 1843 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu", 1844 err, block_start, block_stop); 1845 return err; 1846 } 1847 1848 /* reset start and stop after processing a region */ 1849 block_start = ~0ULL; 1850 } 1851 1852 return blkdev_issue_flush(journal->j_dev); 1853 } 1854 1855 /** 1856 * jbd2_journal_update_sb_errno() - Update error in the journal. 1857 * @journal: The journal to update. 1858 * 1859 * Update a journal's errno. Write updated superblock to disk waiting for IO 1860 * to complete. 1861 */ 1862 void jbd2_journal_update_sb_errno(journal_t *journal) 1863 { 1864 journal_superblock_t *sb = journal->j_superblock; 1865 int errcode; 1866 1867 lock_buffer(journal->j_sb_buffer); 1868 errcode = journal->j_errno; 1869 if (errcode == -ESHUTDOWN) 1870 errcode = 0; 1871 jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode); 1872 sb->s_errno = cpu_to_be32(errcode); 1873 1874 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA); 1875 } 1876 EXPORT_SYMBOL(jbd2_journal_update_sb_errno); 1877 1878 static int journal_revoke_records_per_block(journal_t *journal) 1879 { 1880 int record_size; 1881 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t); 1882 1883 if (jbd2_has_feature_64bit(journal)) 1884 record_size = 8; 1885 else 1886 record_size = 4; 1887 1888 if (jbd2_journal_has_csum_v2or3(journal)) 1889 space -= sizeof(struct jbd2_journal_block_tail); 1890 return space / record_size; 1891 } 1892 1893 /* 1894 * Read the superblock for a given journal, performing initial 1895 * validation of the format. 1896 */ 1897 static int journal_get_superblock(journal_t *journal) 1898 { 1899 struct buffer_head *bh; 1900 journal_superblock_t *sb; 1901 int err; 1902 1903 bh = journal->j_sb_buffer; 1904 1905 J_ASSERT(bh != NULL); 1906 err = bh_read(bh, 0); 1907 if (err < 0) { 1908 printk(KERN_ERR 1909 "JBD2: IO error reading journal superblock\n"); 1910 goto out; 1911 } 1912 1913 if (buffer_verified(bh)) 1914 return 0; 1915 1916 sb = journal->j_superblock; 1917 1918 err = -EINVAL; 1919 1920 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || 1921 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { 1922 printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); 1923 goto out; 1924 } 1925 1926 switch(be32_to_cpu(sb->s_header.h_blocktype)) { 1927 case JBD2_SUPERBLOCK_V1: 1928 journal->j_format_version = 1; 1929 break; 1930 case JBD2_SUPERBLOCK_V2: 1931 journal->j_format_version = 2; 1932 break; 1933 default: 1934 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); 1935 goto out; 1936 } 1937 1938 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len) 1939 journal->j_total_len = be32_to_cpu(sb->s_maxlen); 1940 else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) { 1941 printk(KERN_WARNING "JBD2: journal file too short\n"); 1942 goto out; 1943 } 1944 1945 if (be32_to_cpu(sb->s_first) == 0 || 1946 be32_to_cpu(sb->s_first) >= journal->j_total_len) { 1947 printk(KERN_WARNING 1948 "JBD2: Invalid start block of journal: %u\n", 1949 be32_to_cpu(sb->s_first)); 1950 goto out; 1951 } 1952 1953 if (jbd2_has_feature_csum2(journal) && 1954 jbd2_has_feature_csum3(journal)) { 1955 /* Can't have checksum v2 and v3 at the same time! */ 1956 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " 1957 "at the same time!\n"); 1958 goto out; 1959 } 1960 1961 if (jbd2_journal_has_csum_v2or3_feature(journal) && 1962 jbd2_has_feature_checksum(journal)) { 1963 /* Can't have checksum v1 and v2 on at the same time! */ 1964 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " 1965 "at the same time!\n"); 1966 goto out; 1967 } 1968 1969 if (!jbd2_verify_csum_type(journal, sb)) { 1970 printk(KERN_ERR "JBD2: Unknown checksum type\n"); 1971 goto out; 1972 } 1973 1974 /* Load the checksum driver */ 1975 if (jbd2_journal_has_csum_v2or3_feature(journal)) { 1976 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 1977 if (IS_ERR(journal->j_chksum_driver)) { 1978 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); 1979 err = PTR_ERR(journal->j_chksum_driver); 1980 journal->j_chksum_driver = NULL; 1981 goto out; 1982 } 1983 } 1984 1985 if (jbd2_journal_has_csum_v2or3(journal)) { 1986 /* Check superblock checksum */ 1987 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) { 1988 printk(KERN_ERR "JBD2: journal checksum error\n"); 1989 err = -EFSBADCRC; 1990 goto out; 1991 } 1992 1993 /* Precompute checksum seed for all metadata */ 1994 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, 1995 sizeof(sb->s_uuid)); 1996 } 1997 1998 journal->j_revoke_records_per_block = 1999 journal_revoke_records_per_block(journal); 2000 set_buffer_verified(bh); 2001 2002 return 0; 2003 2004 out: 2005 journal_fail_superblock(journal); 2006 return err; 2007 } 2008 2009 /* 2010 * Load the on-disk journal superblock and read the key fields into the 2011 * journal_t. 2012 */ 2013 2014 static int load_superblock(journal_t *journal) 2015 { 2016 int err; 2017 journal_superblock_t *sb; 2018 int num_fc_blocks; 2019 2020 err = journal_get_superblock(journal); 2021 if (err) 2022 return err; 2023 2024 sb = journal->j_superblock; 2025 2026 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); 2027 journal->j_tail = be32_to_cpu(sb->s_start); 2028 journal->j_first = be32_to_cpu(sb->s_first); 2029 journal->j_errno = be32_to_cpu(sb->s_errno); 2030 journal->j_last = be32_to_cpu(sb->s_maxlen); 2031 2032 if (jbd2_has_feature_fast_commit(journal)) { 2033 journal->j_fc_last = be32_to_cpu(sb->s_maxlen); 2034 num_fc_blocks = jbd2_journal_get_num_fc_blks(sb); 2035 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS) 2036 journal->j_last = journal->j_fc_last - num_fc_blocks; 2037 journal->j_fc_first = journal->j_last + 1; 2038 journal->j_fc_off = 0; 2039 } 2040 2041 return 0; 2042 } 2043 2044 2045 /** 2046 * jbd2_journal_load() - Read journal from disk. 2047 * @journal: Journal to act on. 2048 * 2049 * Given a journal_t structure which tells us which disk blocks contain 2050 * a journal, read the journal from disk to initialise the in-memory 2051 * structures. 2052 */ 2053 int jbd2_journal_load(journal_t *journal) 2054 { 2055 int err; 2056 journal_superblock_t *sb; 2057 2058 err = load_superblock(journal); 2059 if (err) 2060 return err; 2061 2062 sb = journal->j_superblock; 2063 /* If this is a V2 superblock, then we have to check the 2064 * features flags on it. */ 2065 2066 if (journal->j_format_version >= 2) { 2067 if ((sb->s_feature_ro_compat & 2068 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || 2069 (sb->s_feature_incompat & 2070 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { 2071 printk(KERN_WARNING 2072 "JBD2: Unrecognised features on journal\n"); 2073 return -EINVAL; 2074 } 2075 } 2076 2077 /* 2078 * Create a slab for this blocksize 2079 */ 2080 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); 2081 if (err) 2082 return err; 2083 2084 /* Let the recovery code check whether it needs to recover any 2085 * data from the journal. */ 2086 if (jbd2_journal_recover(journal)) 2087 goto recovery_error; 2088 2089 if (journal->j_failed_commit) { 2090 printk(KERN_ERR "JBD2: journal transaction %u on %s " 2091 "is corrupt.\n", journal->j_failed_commit, 2092 journal->j_devname); 2093 return -EFSCORRUPTED; 2094 } 2095 /* 2096 * clear JBD2_ABORT flag initialized in journal_init_common 2097 * here to update log tail information with the newest seq. 2098 */ 2099 journal->j_flags &= ~JBD2_ABORT; 2100 2101 /* OK, we've finished with the dynamic journal bits: 2102 * reinitialise the dynamic contents of the superblock in memory 2103 * and reset them on disk. */ 2104 if (journal_reset(journal)) 2105 goto recovery_error; 2106 2107 journal->j_flags |= JBD2_LOADED; 2108 return 0; 2109 2110 recovery_error: 2111 printk(KERN_WARNING "JBD2: recovery failed\n"); 2112 return -EIO; 2113 } 2114 2115 /** 2116 * jbd2_journal_destroy() - Release a journal_t structure. 2117 * @journal: Journal to act on. 2118 * 2119 * Release a journal_t structure once it is no longer in use by the 2120 * journaled object. 2121 * Return <0 if we couldn't clean up the journal. 2122 */ 2123 int jbd2_journal_destroy(journal_t *journal) 2124 { 2125 int err = 0; 2126 2127 /* Wait for the commit thread to wake up and die. */ 2128 journal_kill_thread(journal); 2129 2130 /* Force a final log commit */ 2131 if (journal->j_running_transaction) 2132 jbd2_journal_commit_transaction(journal); 2133 2134 /* Force any old transactions to disk */ 2135 2136 /* Totally anal locking here... */ 2137 spin_lock(&journal->j_list_lock); 2138 while (journal->j_checkpoint_transactions != NULL) { 2139 spin_unlock(&journal->j_list_lock); 2140 mutex_lock_io(&journal->j_checkpoint_mutex); 2141 err = jbd2_log_do_checkpoint(journal); 2142 mutex_unlock(&journal->j_checkpoint_mutex); 2143 /* 2144 * If checkpointing failed, just free the buffers to avoid 2145 * looping forever 2146 */ 2147 if (err) { 2148 jbd2_journal_destroy_checkpoint(journal); 2149 spin_lock(&journal->j_list_lock); 2150 break; 2151 } 2152 spin_lock(&journal->j_list_lock); 2153 } 2154 2155 J_ASSERT(journal->j_running_transaction == NULL); 2156 J_ASSERT(journal->j_committing_transaction == NULL); 2157 J_ASSERT(journal->j_checkpoint_transactions == NULL); 2158 spin_unlock(&journal->j_list_lock); 2159 2160 /* 2161 * OK, all checkpoint transactions have been checked, now check the 2162 * write out io error flag and abort the journal if some buffer failed 2163 * to write back to the original location, otherwise the filesystem 2164 * may become inconsistent. 2165 */ 2166 if (!is_journal_aborted(journal) && 2167 test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) 2168 jbd2_journal_abort(journal, -EIO); 2169 2170 if (journal->j_sb_buffer) { 2171 if (!is_journal_aborted(journal)) { 2172 mutex_lock_io(&journal->j_checkpoint_mutex); 2173 2174 write_lock(&journal->j_state_lock); 2175 journal->j_tail_sequence = 2176 ++journal->j_transaction_sequence; 2177 write_unlock(&journal->j_state_lock); 2178 2179 jbd2_mark_journal_empty(journal, 2180 REQ_SYNC | REQ_PREFLUSH | REQ_FUA); 2181 mutex_unlock(&journal->j_checkpoint_mutex); 2182 } else 2183 err = -EIO; 2184 brelse(journal->j_sb_buffer); 2185 } 2186 2187 if (journal->j_shrinker.flags & SHRINKER_REGISTERED) { 2188 percpu_counter_destroy(&journal->j_checkpoint_jh_count); 2189 unregister_shrinker(&journal->j_shrinker); 2190 } 2191 if (journal->j_proc_entry) 2192 jbd2_stats_proc_exit(journal); 2193 iput(journal->j_inode); 2194 if (journal->j_revoke) 2195 jbd2_journal_destroy_revoke(journal); 2196 if (journal->j_chksum_driver) 2197 crypto_free_shash(journal->j_chksum_driver); 2198 kfree(journal->j_fc_wbuf); 2199 kfree(journal->j_wbuf); 2200 kfree(journal); 2201 2202 return err; 2203 } 2204 2205 2206 /** 2207 * jbd2_journal_check_used_features() - Check if features specified are used. 2208 * @journal: Journal to check. 2209 * @compat: bitmask of compatible features 2210 * @ro: bitmask of features that force read-only mount 2211 * @incompat: bitmask of incompatible features 2212 * 2213 * Check whether the journal uses all of a given set of 2214 * features. Return true (non-zero) if it does. 2215 **/ 2216 2217 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat, 2218 unsigned long ro, unsigned long incompat) 2219 { 2220 journal_superblock_t *sb; 2221 2222 if (!compat && !ro && !incompat) 2223 return 1; 2224 /* Load journal superblock if it is not loaded yet. */ 2225 if (journal->j_format_version == 0 && 2226 journal_get_superblock(journal) != 0) 2227 return 0; 2228 if (journal->j_format_version == 1) 2229 return 0; 2230 2231 sb = journal->j_superblock; 2232 2233 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && 2234 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && 2235 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) 2236 return 1; 2237 2238 return 0; 2239 } 2240 2241 /** 2242 * jbd2_journal_check_available_features() - Check feature set in journalling layer 2243 * @journal: Journal to check. 2244 * @compat: bitmask of compatible features 2245 * @ro: bitmask of features that force read-only mount 2246 * @incompat: bitmask of incompatible features 2247 * 2248 * Check whether the journaling code supports the use of 2249 * all of a given set of features on this journal. Return true 2250 * (non-zero) if it can. */ 2251 2252 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat, 2253 unsigned long ro, unsigned long incompat) 2254 { 2255 if (!compat && !ro && !incompat) 2256 return 1; 2257 2258 /* We can support any known requested features iff the 2259 * superblock is in version 2. Otherwise we fail to support any 2260 * extended sb features. */ 2261 2262 if (journal->j_format_version != 2) 2263 return 0; 2264 2265 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && 2266 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && 2267 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) 2268 return 1; 2269 2270 return 0; 2271 } 2272 2273 static int 2274 jbd2_journal_initialize_fast_commit(journal_t *journal) 2275 { 2276 journal_superblock_t *sb = journal->j_superblock; 2277 unsigned long long num_fc_blks; 2278 2279 num_fc_blks = jbd2_journal_get_num_fc_blks(sb); 2280 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS) 2281 return -ENOSPC; 2282 2283 /* Are we called twice? */ 2284 WARN_ON(journal->j_fc_wbuf != NULL); 2285 journal->j_fc_wbuf = kmalloc_array(num_fc_blks, 2286 sizeof(struct buffer_head *), GFP_KERNEL); 2287 if (!journal->j_fc_wbuf) 2288 return -ENOMEM; 2289 2290 journal->j_fc_wbufsize = num_fc_blks; 2291 journal->j_fc_last = journal->j_last; 2292 journal->j_last = journal->j_fc_last - num_fc_blks; 2293 journal->j_fc_first = journal->j_last + 1; 2294 journal->j_fc_off = 0; 2295 journal->j_free = journal->j_last - journal->j_first; 2296 journal->j_max_transaction_buffers = 2297 jbd2_journal_get_max_txn_bufs(journal); 2298 2299 return 0; 2300 } 2301 2302 /** 2303 * jbd2_journal_set_features() - Mark a given journal feature in the superblock 2304 * @journal: Journal to act on. 2305 * @compat: bitmask of compatible features 2306 * @ro: bitmask of features that force read-only mount 2307 * @incompat: bitmask of incompatible features 2308 * 2309 * Mark a given journal feature as present on the 2310 * superblock. Returns true if the requested features could be set. 2311 * 2312 */ 2313 2314 int jbd2_journal_set_features(journal_t *journal, unsigned long compat, 2315 unsigned long ro, unsigned long incompat) 2316 { 2317 #define INCOMPAT_FEATURE_ON(f) \ 2318 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) 2319 #define COMPAT_FEATURE_ON(f) \ 2320 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) 2321 journal_superblock_t *sb; 2322 2323 if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) 2324 return 1; 2325 2326 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) 2327 return 0; 2328 2329 /* If enabling v2 checksums, turn on v3 instead */ 2330 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { 2331 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; 2332 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; 2333 } 2334 2335 /* Asking for checksumming v3 and v1? Only give them v3. */ 2336 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && 2337 compat & JBD2_FEATURE_COMPAT_CHECKSUM) 2338 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; 2339 2340 jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", 2341 compat, ro, incompat); 2342 2343 sb = journal->j_superblock; 2344 2345 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) { 2346 if (jbd2_journal_initialize_fast_commit(journal)) { 2347 pr_err("JBD2: Cannot enable fast commits.\n"); 2348 return 0; 2349 } 2350 } 2351 2352 /* Load the checksum driver if necessary */ 2353 if ((journal->j_chksum_driver == NULL) && 2354 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 2355 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 2356 if (IS_ERR(journal->j_chksum_driver)) { 2357 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); 2358 journal->j_chksum_driver = NULL; 2359 return 0; 2360 } 2361 /* Precompute checksum seed for all metadata */ 2362 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, 2363 sizeof(sb->s_uuid)); 2364 } 2365 2366 lock_buffer(journal->j_sb_buffer); 2367 2368 /* If enabling v3 checksums, update superblock */ 2369 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 2370 sb->s_checksum_type = JBD2_CRC32C_CHKSUM; 2371 sb->s_feature_compat &= 2372 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); 2373 } 2374 2375 /* If enabling v1 checksums, downgrade superblock */ 2376 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) 2377 sb->s_feature_incompat &= 2378 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | 2379 JBD2_FEATURE_INCOMPAT_CSUM_V3); 2380 2381 sb->s_feature_compat |= cpu_to_be32(compat); 2382 sb->s_feature_ro_compat |= cpu_to_be32(ro); 2383 sb->s_feature_incompat |= cpu_to_be32(incompat); 2384 unlock_buffer(journal->j_sb_buffer); 2385 journal->j_revoke_records_per_block = 2386 journal_revoke_records_per_block(journal); 2387 2388 return 1; 2389 #undef COMPAT_FEATURE_ON 2390 #undef INCOMPAT_FEATURE_ON 2391 } 2392 2393 /* 2394 * jbd2_journal_clear_features() - Clear a given journal feature in the 2395 * superblock 2396 * @journal: Journal to act on. 2397 * @compat: bitmask of compatible features 2398 * @ro: bitmask of features that force read-only mount 2399 * @incompat: bitmask of incompatible features 2400 * 2401 * Clear a given journal feature as present on the 2402 * superblock. 2403 */ 2404 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, 2405 unsigned long ro, unsigned long incompat) 2406 { 2407 journal_superblock_t *sb; 2408 2409 jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", 2410 compat, ro, incompat); 2411 2412 sb = journal->j_superblock; 2413 2414 sb->s_feature_compat &= ~cpu_to_be32(compat); 2415 sb->s_feature_ro_compat &= ~cpu_to_be32(ro); 2416 sb->s_feature_incompat &= ~cpu_to_be32(incompat); 2417 journal->j_revoke_records_per_block = 2418 journal_revoke_records_per_block(journal); 2419 } 2420 EXPORT_SYMBOL(jbd2_journal_clear_features); 2421 2422 /** 2423 * jbd2_journal_flush() - Flush journal 2424 * @journal: Journal to act on. 2425 * @flags: optional operation on the journal blocks after the flush (see below) 2426 * 2427 * Flush all data for a given journal to disk and empty the journal. 2428 * Filesystems can use this when remounting readonly to ensure that 2429 * recovery does not need to happen on remount. Optionally, a discard or zeroout 2430 * can be issued on the journal blocks after flushing. 2431 * 2432 * flags: 2433 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks 2434 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks 2435 */ 2436 int jbd2_journal_flush(journal_t *journal, unsigned int flags) 2437 { 2438 int err = 0; 2439 transaction_t *transaction = NULL; 2440 2441 write_lock(&journal->j_state_lock); 2442 2443 /* Force everything buffered to the log... */ 2444 if (journal->j_running_transaction) { 2445 transaction = journal->j_running_transaction; 2446 __jbd2_log_start_commit(journal, transaction->t_tid); 2447 } else if (journal->j_committing_transaction) 2448 transaction = journal->j_committing_transaction; 2449 2450 /* Wait for the log commit to complete... */ 2451 if (transaction) { 2452 tid_t tid = transaction->t_tid; 2453 2454 write_unlock(&journal->j_state_lock); 2455 jbd2_log_wait_commit(journal, tid); 2456 } else { 2457 write_unlock(&journal->j_state_lock); 2458 } 2459 2460 /* ...and flush everything in the log out to disk. */ 2461 spin_lock(&journal->j_list_lock); 2462 while (!err && journal->j_checkpoint_transactions != NULL) { 2463 spin_unlock(&journal->j_list_lock); 2464 mutex_lock_io(&journal->j_checkpoint_mutex); 2465 err = jbd2_log_do_checkpoint(journal); 2466 mutex_unlock(&journal->j_checkpoint_mutex); 2467 spin_lock(&journal->j_list_lock); 2468 } 2469 spin_unlock(&journal->j_list_lock); 2470 2471 if (is_journal_aborted(journal)) 2472 return -EIO; 2473 2474 mutex_lock_io(&journal->j_checkpoint_mutex); 2475 if (!err) { 2476 err = jbd2_cleanup_journal_tail(journal); 2477 if (err < 0) { 2478 mutex_unlock(&journal->j_checkpoint_mutex); 2479 goto out; 2480 } 2481 err = 0; 2482 } 2483 2484 /* Finally, mark the journal as really needing no recovery. 2485 * This sets s_start==0 in the underlying superblock, which is 2486 * the magic code for a fully-recovered superblock. Any future 2487 * commits of data to the journal will restore the current 2488 * s_start value. */ 2489 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); 2490 2491 if (flags) 2492 err = __jbd2_journal_erase(journal, flags); 2493 2494 mutex_unlock(&journal->j_checkpoint_mutex); 2495 write_lock(&journal->j_state_lock); 2496 J_ASSERT(!journal->j_running_transaction); 2497 J_ASSERT(!journal->j_committing_transaction); 2498 J_ASSERT(!journal->j_checkpoint_transactions); 2499 J_ASSERT(journal->j_head == journal->j_tail); 2500 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); 2501 write_unlock(&journal->j_state_lock); 2502 out: 2503 return err; 2504 } 2505 2506 /** 2507 * jbd2_journal_wipe() - Wipe journal contents 2508 * @journal: Journal to act on. 2509 * @write: flag (see below) 2510 * 2511 * Wipe out all of the contents of a journal, safely. This will produce 2512 * a warning if the journal contains any valid recovery information. 2513 * Must be called between journal_init_*() and jbd2_journal_load(). 2514 * 2515 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise 2516 * we merely suppress recovery. 2517 */ 2518 2519 int jbd2_journal_wipe(journal_t *journal, int write) 2520 { 2521 int err = 0; 2522 2523 J_ASSERT (!(journal->j_flags & JBD2_LOADED)); 2524 2525 err = load_superblock(journal); 2526 if (err) 2527 return err; 2528 2529 if (!journal->j_tail) 2530 goto no_recovery; 2531 2532 printk(KERN_WARNING "JBD2: %s recovery information on journal\n", 2533 write ? "Clearing" : "Ignoring"); 2534 2535 err = jbd2_journal_skip_recovery(journal); 2536 if (write) { 2537 /* Lock to make assertions happy... */ 2538 mutex_lock_io(&journal->j_checkpoint_mutex); 2539 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); 2540 mutex_unlock(&journal->j_checkpoint_mutex); 2541 } 2542 2543 no_recovery: 2544 return err; 2545 } 2546 2547 /** 2548 * jbd2_journal_abort () - Shutdown the journal immediately. 2549 * @journal: the journal to shutdown. 2550 * @errno: an error number to record in the journal indicating 2551 * the reason for the shutdown. 2552 * 2553 * Perform a complete, immediate shutdown of the ENTIRE 2554 * journal (not of a single transaction). This operation cannot be 2555 * undone without closing and reopening the journal. 2556 * 2557 * The jbd2_journal_abort function is intended to support higher level error 2558 * recovery mechanisms such as the ext2/ext3 remount-readonly error 2559 * mode. 2560 * 2561 * Journal abort has very specific semantics. Any existing dirty, 2562 * unjournaled buffers in the main filesystem will still be written to 2563 * disk by bdflush, but the journaling mechanism will be suspended 2564 * immediately and no further transaction commits will be honoured. 2565 * 2566 * Any dirty, journaled buffers will be written back to disk without 2567 * hitting the journal. Atomicity cannot be guaranteed on an aborted 2568 * filesystem, but we _do_ attempt to leave as much data as possible 2569 * behind for fsck to use for cleanup. 2570 * 2571 * Any attempt to get a new transaction handle on a journal which is in 2572 * ABORT state will just result in an -EROFS error return. A 2573 * jbd2_journal_stop on an existing handle will return -EIO if we have 2574 * entered abort state during the update. 2575 * 2576 * Recursive transactions are not disturbed by journal abort until the 2577 * final jbd2_journal_stop, which will receive the -EIO error. 2578 * 2579 * Finally, the jbd2_journal_abort call allows the caller to supply an errno 2580 * which will be recorded (if possible) in the journal superblock. This 2581 * allows a client to record failure conditions in the middle of a 2582 * transaction without having to complete the transaction to record the 2583 * failure to disk. ext3_error, for example, now uses this 2584 * functionality. 2585 * 2586 */ 2587 2588 void jbd2_journal_abort(journal_t *journal, int errno) 2589 { 2590 transaction_t *transaction; 2591 2592 /* 2593 * Lock the aborting procedure until everything is done, this avoid 2594 * races between filesystem's error handling flow (e.g. ext4_abort()), 2595 * ensure panic after the error info is written into journal's 2596 * superblock. 2597 */ 2598 mutex_lock(&journal->j_abort_mutex); 2599 /* 2600 * ESHUTDOWN always takes precedence because a file system check 2601 * caused by any other journal abort error is not required after 2602 * a shutdown triggered. 2603 */ 2604 write_lock(&journal->j_state_lock); 2605 if (journal->j_flags & JBD2_ABORT) { 2606 int old_errno = journal->j_errno; 2607 2608 write_unlock(&journal->j_state_lock); 2609 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) { 2610 journal->j_errno = errno; 2611 jbd2_journal_update_sb_errno(journal); 2612 } 2613 mutex_unlock(&journal->j_abort_mutex); 2614 return; 2615 } 2616 2617 /* 2618 * Mark the abort as occurred and start current running transaction 2619 * to release all journaled buffer. 2620 */ 2621 pr_err("Aborting journal on device %s.\n", journal->j_devname); 2622 2623 journal->j_flags |= JBD2_ABORT; 2624 journal->j_errno = errno; 2625 transaction = journal->j_running_transaction; 2626 if (transaction) 2627 __jbd2_log_start_commit(journal, transaction->t_tid); 2628 write_unlock(&journal->j_state_lock); 2629 2630 /* 2631 * Record errno to the journal super block, so that fsck and jbd2 2632 * layer could realise that a filesystem check is needed. 2633 */ 2634 jbd2_journal_update_sb_errno(journal); 2635 mutex_unlock(&journal->j_abort_mutex); 2636 } 2637 2638 /** 2639 * jbd2_journal_errno() - returns the journal's error state. 2640 * @journal: journal to examine. 2641 * 2642 * This is the errno number set with jbd2_journal_abort(), the last 2643 * time the journal was mounted - if the journal was stopped 2644 * without calling abort this will be 0. 2645 * 2646 * If the journal has been aborted on this mount time -EROFS will 2647 * be returned. 2648 */ 2649 int jbd2_journal_errno(journal_t *journal) 2650 { 2651 int err; 2652 2653 read_lock(&journal->j_state_lock); 2654 if (journal->j_flags & JBD2_ABORT) 2655 err = -EROFS; 2656 else 2657 err = journal->j_errno; 2658 read_unlock(&journal->j_state_lock); 2659 return err; 2660 } 2661 2662 /** 2663 * jbd2_journal_clear_err() - clears the journal's error state 2664 * @journal: journal to act on. 2665 * 2666 * An error must be cleared or acked to take a FS out of readonly 2667 * mode. 2668 */ 2669 int jbd2_journal_clear_err(journal_t *journal) 2670 { 2671 int err = 0; 2672 2673 write_lock(&journal->j_state_lock); 2674 if (journal->j_flags & JBD2_ABORT) 2675 err = -EROFS; 2676 else 2677 journal->j_errno = 0; 2678 write_unlock(&journal->j_state_lock); 2679 return err; 2680 } 2681 2682 /** 2683 * jbd2_journal_ack_err() - Ack journal err. 2684 * @journal: journal to act on. 2685 * 2686 * An error must be cleared or acked to take a FS out of readonly 2687 * mode. 2688 */ 2689 void jbd2_journal_ack_err(journal_t *journal) 2690 { 2691 write_lock(&journal->j_state_lock); 2692 if (journal->j_errno) 2693 journal->j_flags |= JBD2_ACK_ERR; 2694 write_unlock(&journal->j_state_lock); 2695 } 2696 2697 int jbd2_journal_blocks_per_page(struct inode *inode) 2698 { 2699 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); 2700 } 2701 2702 /* 2703 * helper functions to deal with 32 or 64bit block numbers. 2704 */ 2705 size_t journal_tag_bytes(journal_t *journal) 2706 { 2707 size_t sz; 2708 2709 if (jbd2_has_feature_csum3(journal)) 2710 return sizeof(journal_block_tag3_t); 2711 2712 sz = sizeof(journal_block_tag_t); 2713 2714 if (jbd2_has_feature_csum2(journal)) 2715 sz += sizeof(__u16); 2716 2717 if (jbd2_has_feature_64bit(journal)) 2718 return sz; 2719 else 2720 return sz - sizeof(__u32); 2721 } 2722 2723 /* 2724 * JBD memory management 2725 * 2726 * These functions are used to allocate block-sized chunks of memory 2727 * used for making copies of buffer_head data. Very often it will be 2728 * page-sized chunks of data, but sometimes it will be in 2729 * sub-page-size chunks. (For example, 16k pages on Power systems 2730 * with a 4k block file system.) For blocks smaller than a page, we 2731 * use a SLAB allocator. There are slab caches for each block size, 2732 * which are allocated at mount time, if necessary, and we only free 2733 * (all of) the slab caches when/if the jbd2 module is unloaded. For 2734 * this reason we don't need to a mutex to protect access to 2735 * jbd2_slab[] allocating or releasing memory; only in 2736 * jbd2_journal_create_slab(). 2737 */ 2738 #define JBD2_MAX_SLABS 8 2739 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; 2740 2741 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { 2742 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", 2743 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" 2744 }; 2745 2746 2747 static void jbd2_journal_destroy_slabs(void) 2748 { 2749 int i; 2750 2751 for (i = 0; i < JBD2_MAX_SLABS; i++) { 2752 kmem_cache_destroy(jbd2_slab[i]); 2753 jbd2_slab[i] = NULL; 2754 } 2755 } 2756 2757 static int jbd2_journal_create_slab(size_t size) 2758 { 2759 static DEFINE_MUTEX(jbd2_slab_create_mutex); 2760 int i = order_base_2(size) - 10; 2761 size_t slab_size; 2762 2763 if (size == PAGE_SIZE) 2764 return 0; 2765 2766 if (i >= JBD2_MAX_SLABS) 2767 return -EINVAL; 2768 2769 if (unlikely(i < 0)) 2770 i = 0; 2771 mutex_lock(&jbd2_slab_create_mutex); 2772 if (jbd2_slab[i]) { 2773 mutex_unlock(&jbd2_slab_create_mutex); 2774 return 0; /* Already created */ 2775 } 2776 2777 slab_size = 1 << (i+10); 2778 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, 2779 slab_size, 0, NULL); 2780 mutex_unlock(&jbd2_slab_create_mutex); 2781 if (!jbd2_slab[i]) { 2782 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); 2783 return -ENOMEM; 2784 } 2785 return 0; 2786 } 2787 2788 static struct kmem_cache *get_slab(size_t size) 2789 { 2790 int i = order_base_2(size) - 10; 2791 2792 BUG_ON(i >= JBD2_MAX_SLABS); 2793 if (unlikely(i < 0)) 2794 i = 0; 2795 BUG_ON(jbd2_slab[i] == NULL); 2796 return jbd2_slab[i]; 2797 } 2798 2799 void *jbd2_alloc(size_t size, gfp_t flags) 2800 { 2801 void *ptr; 2802 2803 BUG_ON(size & (size-1)); /* Must be a power of 2 */ 2804 2805 if (size < PAGE_SIZE) 2806 ptr = kmem_cache_alloc(get_slab(size), flags); 2807 else 2808 ptr = (void *)__get_free_pages(flags, get_order(size)); 2809 2810 /* Check alignment; SLUB has gotten this wrong in the past, 2811 * and this can lead to user data corruption! */ 2812 BUG_ON(((unsigned long) ptr) & (size-1)); 2813 2814 return ptr; 2815 } 2816 2817 void jbd2_free(void *ptr, size_t size) 2818 { 2819 if (size < PAGE_SIZE) 2820 kmem_cache_free(get_slab(size), ptr); 2821 else 2822 free_pages((unsigned long)ptr, get_order(size)); 2823 }; 2824 2825 /* 2826 * Journal_head storage management 2827 */ 2828 static struct kmem_cache *jbd2_journal_head_cache; 2829 #ifdef CONFIG_JBD2_DEBUG 2830 static atomic_t nr_journal_heads = ATOMIC_INIT(0); 2831 #endif 2832 2833 static int __init jbd2_journal_init_journal_head_cache(void) 2834 { 2835 J_ASSERT(!jbd2_journal_head_cache); 2836 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", 2837 sizeof(struct journal_head), 2838 0, /* offset */ 2839 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU, 2840 NULL); /* ctor */ 2841 if (!jbd2_journal_head_cache) { 2842 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); 2843 return -ENOMEM; 2844 } 2845 return 0; 2846 } 2847 2848 static void jbd2_journal_destroy_journal_head_cache(void) 2849 { 2850 kmem_cache_destroy(jbd2_journal_head_cache); 2851 jbd2_journal_head_cache = NULL; 2852 } 2853 2854 /* 2855 * journal_head splicing and dicing 2856 */ 2857 static struct journal_head *journal_alloc_journal_head(void) 2858 { 2859 struct journal_head *ret; 2860 2861 #ifdef CONFIG_JBD2_DEBUG 2862 atomic_inc(&nr_journal_heads); 2863 #endif 2864 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); 2865 if (!ret) { 2866 jbd2_debug(1, "out of memory for journal_head\n"); 2867 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); 2868 ret = kmem_cache_zalloc(jbd2_journal_head_cache, 2869 GFP_NOFS | __GFP_NOFAIL); 2870 } 2871 if (ret) 2872 spin_lock_init(&ret->b_state_lock); 2873 return ret; 2874 } 2875 2876 static void journal_free_journal_head(struct journal_head *jh) 2877 { 2878 #ifdef CONFIG_JBD2_DEBUG 2879 atomic_dec(&nr_journal_heads); 2880 memset(jh, JBD2_POISON_FREE, sizeof(*jh)); 2881 #endif 2882 kmem_cache_free(jbd2_journal_head_cache, jh); 2883 } 2884 2885 /* 2886 * A journal_head is attached to a buffer_head whenever JBD has an 2887 * interest in the buffer. 2888 * 2889 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit 2890 * is set. This bit is tested in core kernel code where we need to take 2891 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable 2892 * there. 2893 * 2894 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. 2895 * 2896 * When a buffer has its BH_JBD bit set it is immune from being released by 2897 * core kernel code, mainly via ->b_count. 2898 * 2899 * A journal_head is detached from its buffer_head when the journal_head's 2900 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint 2901 * transaction (b_cp_transaction) hold their references to b_jcount. 2902 * 2903 * Various places in the kernel want to attach a journal_head to a buffer_head 2904 * _before_ attaching the journal_head to a transaction. To protect the 2905 * journal_head in this situation, jbd2_journal_add_journal_head elevates the 2906 * journal_head's b_jcount refcount by one. The caller must call 2907 * jbd2_journal_put_journal_head() to undo this. 2908 * 2909 * So the typical usage would be: 2910 * 2911 * (Attach a journal_head if needed. Increments b_jcount) 2912 * struct journal_head *jh = jbd2_journal_add_journal_head(bh); 2913 * ... 2914 * (Get another reference for transaction) 2915 * jbd2_journal_grab_journal_head(bh); 2916 * jh->b_transaction = xxx; 2917 * (Put original reference) 2918 * jbd2_journal_put_journal_head(jh); 2919 */ 2920 2921 /* 2922 * Give a buffer_head a journal_head. 2923 * 2924 * May sleep. 2925 */ 2926 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) 2927 { 2928 struct journal_head *jh; 2929 struct journal_head *new_jh = NULL; 2930 2931 repeat: 2932 if (!buffer_jbd(bh)) 2933 new_jh = journal_alloc_journal_head(); 2934 2935 jbd_lock_bh_journal_head(bh); 2936 if (buffer_jbd(bh)) { 2937 jh = bh2jh(bh); 2938 } else { 2939 J_ASSERT_BH(bh, 2940 (atomic_read(&bh->b_count) > 0) || 2941 (bh->b_folio && bh->b_folio->mapping)); 2942 2943 if (!new_jh) { 2944 jbd_unlock_bh_journal_head(bh); 2945 goto repeat; 2946 } 2947 2948 jh = new_jh; 2949 new_jh = NULL; /* We consumed it */ 2950 set_buffer_jbd(bh); 2951 bh->b_private = jh; 2952 jh->b_bh = bh; 2953 get_bh(bh); 2954 BUFFER_TRACE(bh, "added journal_head"); 2955 } 2956 jh->b_jcount++; 2957 jbd_unlock_bh_journal_head(bh); 2958 if (new_jh) 2959 journal_free_journal_head(new_jh); 2960 return bh->b_private; 2961 } 2962 2963 /* 2964 * Grab a ref against this buffer_head's journal_head. If it ended up not 2965 * having a journal_head, return NULL 2966 */ 2967 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) 2968 { 2969 struct journal_head *jh = NULL; 2970 2971 jbd_lock_bh_journal_head(bh); 2972 if (buffer_jbd(bh)) { 2973 jh = bh2jh(bh); 2974 jh->b_jcount++; 2975 } 2976 jbd_unlock_bh_journal_head(bh); 2977 return jh; 2978 } 2979 EXPORT_SYMBOL(jbd2_journal_grab_journal_head); 2980 2981 static void __journal_remove_journal_head(struct buffer_head *bh) 2982 { 2983 struct journal_head *jh = bh2jh(bh); 2984 2985 J_ASSERT_JH(jh, jh->b_transaction == NULL); 2986 J_ASSERT_JH(jh, jh->b_next_transaction == NULL); 2987 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); 2988 J_ASSERT_JH(jh, jh->b_jlist == BJ_None); 2989 J_ASSERT_BH(bh, buffer_jbd(bh)); 2990 J_ASSERT_BH(bh, jh2bh(jh) == bh); 2991 BUFFER_TRACE(bh, "remove journal_head"); 2992 2993 /* Unlink before dropping the lock */ 2994 bh->b_private = NULL; 2995 jh->b_bh = NULL; /* debug, really */ 2996 clear_buffer_jbd(bh); 2997 } 2998 2999 static void journal_release_journal_head(struct journal_head *jh, size_t b_size) 3000 { 3001 if (jh->b_frozen_data) { 3002 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); 3003 jbd2_free(jh->b_frozen_data, b_size); 3004 } 3005 if (jh->b_committed_data) { 3006 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); 3007 jbd2_free(jh->b_committed_data, b_size); 3008 } 3009 journal_free_journal_head(jh); 3010 } 3011 3012 /* 3013 * Drop a reference on the passed journal_head. If it fell to zero then 3014 * release the journal_head from the buffer_head. 3015 */ 3016 void jbd2_journal_put_journal_head(struct journal_head *jh) 3017 { 3018 struct buffer_head *bh = jh2bh(jh); 3019 3020 jbd_lock_bh_journal_head(bh); 3021 J_ASSERT_JH(jh, jh->b_jcount > 0); 3022 --jh->b_jcount; 3023 if (!jh->b_jcount) { 3024 __journal_remove_journal_head(bh); 3025 jbd_unlock_bh_journal_head(bh); 3026 journal_release_journal_head(jh, bh->b_size); 3027 __brelse(bh); 3028 } else { 3029 jbd_unlock_bh_journal_head(bh); 3030 } 3031 } 3032 EXPORT_SYMBOL(jbd2_journal_put_journal_head); 3033 3034 /* 3035 * Initialize jbd inode head 3036 */ 3037 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) 3038 { 3039 jinode->i_transaction = NULL; 3040 jinode->i_next_transaction = NULL; 3041 jinode->i_vfs_inode = inode; 3042 jinode->i_flags = 0; 3043 jinode->i_dirty_start = 0; 3044 jinode->i_dirty_end = 0; 3045 INIT_LIST_HEAD(&jinode->i_list); 3046 } 3047 3048 /* 3049 * Function to be called before we start removing inode from memory (i.e., 3050 * clear_inode() is a fine place to be called from). It removes inode from 3051 * transaction's lists. 3052 */ 3053 void jbd2_journal_release_jbd_inode(journal_t *journal, 3054 struct jbd2_inode *jinode) 3055 { 3056 if (!journal) 3057 return; 3058 restart: 3059 spin_lock(&journal->j_list_lock); 3060 /* Is commit writing out inode - we have to wait */ 3061 if (jinode->i_flags & JI_COMMIT_RUNNING) { 3062 wait_queue_head_t *wq; 3063 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); 3064 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); 3065 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 3066 spin_unlock(&journal->j_list_lock); 3067 schedule(); 3068 finish_wait(wq, &wait.wq_entry); 3069 goto restart; 3070 } 3071 3072 if (jinode->i_transaction) { 3073 list_del(&jinode->i_list); 3074 jinode->i_transaction = NULL; 3075 } 3076 spin_unlock(&journal->j_list_lock); 3077 } 3078 3079 3080 #ifdef CONFIG_PROC_FS 3081 3082 #define JBD2_STATS_PROC_NAME "fs/jbd2" 3083 3084 static void __init jbd2_create_jbd_stats_proc_entry(void) 3085 { 3086 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); 3087 } 3088 3089 static void __exit jbd2_remove_jbd_stats_proc_entry(void) 3090 { 3091 if (proc_jbd2_stats) 3092 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); 3093 } 3094 3095 #else 3096 3097 #define jbd2_create_jbd_stats_proc_entry() do {} while (0) 3098 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) 3099 3100 #endif 3101 3102 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; 3103 3104 static int __init jbd2_journal_init_inode_cache(void) 3105 { 3106 J_ASSERT(!jbd2_inode_cache); 3107 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); 3108 if (!jbd2_inode_cache) { 3109 pr_emerg("JBD2: failed to create inode cache\n"); 3110 return -ENOMEM; 3111 } 3112 return 0; 3113 } 3114 3115 static int __init jbd2_journal_init_handle_cache(void) 3116 { 3117 J_ASSERT(!jbd2_handle_cache); 3118 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); 3119 if (!jbd2_handle_cache) { 3120 printk(KERN_EMERG "JBD2: failed to create handle cache\n"); 3121 return -ENOMEM; 3122 } 3123 return 0; 3124 } 3125 3126 static void jbd2_journal_destroy_inode_cache(void) 3127 { 3128 kmem_cache_destroy(jbd2_inode_cache); 3129 jbd2_inode_cache = NULL; 3130 } 3131 3132 static void jbd2_journal_destroy_handle_cache(void) 3133 { 3134 kmem_cache_destroy(jbd2_handle_cache); 3135 jbd2_handle_cache = NULL; 3136 } 3137 3138 /* 3139 * Module startup and shutdown 3140 */ 3141 3142 static int __init journal_init_caches(void) 3143 { 3144 int ret; 3145 3146 ret = jbd2_journal_init_revoke_record_cache(); 3147 if (ret == 0) 3148 ret = jbd2_journal_init_revoke_table_cache(); 3149 if (ret == 0) 3150 ret = jbd2_journal_init_journal_head_cache(); 3151 if (ret == 0) 3152 ret = jbd2_journal_init_handle_cache(); 3153 if (ret == 0) 3154 ret = jbd2_journal_init_inode_cache(); 3155 if (ret == 0) 3156 ret = jbd2_journal_init_transaction_cache(); 3157 return ret; 3158 } 3159 3160 static void jbd2_journal_destroy_caches(void) 3161 { 3162 jbd2_journal_destroy_revoke_record_cache(); 3163 jbd2_journal_destroy_revoke_table_cache(); 3164 jbd2_journal_destroy_journal_head_cache(); 3165 jbd2_journal_destroy_handle_cache(); 3166 jbd2_journal_destroy_inode_cache(); 3167 jbd2_journal_destroy_transaction_cache(); 3168 jbd2_journal_destroy_slabs(); 3169 } 3170 3171 static int __init journal_init(void) 3172 { 3173 int ret; 3174 3175 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); 3176 3177 ret = journal_init_caches(); 3178 if (ret == 0) { 3179 jbd2_create_jbd_stats_proc_entry(); 3180 } else { 3181 jbd2_journal_destroy_caches(); 3182 } 3183 return ret; 3184 } 3185 3186 static void __exit journal_exit(void) 3187 { 3188 #ifdef CONFIG_JBD2_DEBUG 3189 int n = atomic_read(&nr_journal_heads); 3190 if (n) 3191 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); 3192 #endif 3193 jbd2_remove_jbd_stats_proc_entry(); 3194 jbd2_journal_destroy_caches(); 3195 } 3196 3197 MODULE_LICENSE("GPL"); 3198 module_init(journal_init); 3199 module_exit(journal_exit); 3200 3201