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 ushort jbd2_journal_enable_debug __read_mostly; 53 EXPORT_SYMBOL(jbd2_journal_enable_debug); 54 55 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644); 56 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2"); 57 #endif 58 59 EXPORT_SYMBOL(jbd2_journal_extend); 60 EXPORT_SYMBOL(jbd2_journal_stop); 61 EXPORT_SYMBOL(jbd2_journal_lock_updates); 62 EXPORT_SYMBOL(jbd2_journal_unlock_updates); 63 EXPORT_SYMBOL(jbd2_journal_get_write_access); 64 EXPORT_SYMBOL(jbd2_journal_get_create_access); 65 EXPORT_SYMBOL(jbd2_journal_get_undo_access); 66 EXPORT_SYMBOL(jbd2_journal_set_triggers); 67 EXPORT_SYMBOL(jbd2_journal_dirty_metadata); 68 EXPORT_SYMBOL(jbd2_journal_forget); 69 EXPORT_SYMBOL(jbd2_journal_flush); 70 EXPORT_SYMBOL(jbd2_journal_revoke); 71 72 EXPORT_SYMBOL(jbd2_journal_init_dev); 73 EXPORT_SYMBOL(jbd2_journal_init_inode); 74 EXPORT_SYMBOL(jbd2_journal_check_used_features); 75 EXPORT_SYMBOL(jbd2_journal_check_available_features); 76 EXPORT_SYMBOL(jbd2_journal_set_features); 77 EXPORT_SYMBOL(jbd2_journal_load); 78 EXPORT_SYMBOL(jbd2_journal_destroy); 79 EXPORT_SYMBOL(jbd2_journal_abort); 80 EXPORT_SYMBOL(jbd2_journal_errno); 81 EXPORT_SYMBOL(jbd2_journal_ack_err); 82 EXPORT_SYMBOL(jbd2_journal_clear_err); 83 EXPORT_SYMBOL(jbd2_log_wait_commit); 84 EXPORT_SYMBOL(jbd2_log_start_commit); 85 EXPORT_SYMBOL(jbd2_journal_start_commit); 86 EXPORT_SYMBOL(jbd2_journal_force_commit_nested); 87 EXPORT_SYMBOL(jbd2_journal_wipe); 88 EXPORT_SYMBOL(jbd2_journal_blocks_per_page); 89 EXPORT_SYMBOL(jbd2_journal_invalidate_folio); 90 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); 91 EXPORT_SYMBOL(jbd2_journal_force_commit); 92 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write); 93 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait); 94 EXPORT_SYMBOL(jbd2_journal_submit_inode_data_buffers); 95 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers); 96 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); 97 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); 98 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); 99 EXPORT_SYMBOL(jbd2_inode_cache); 100 101 static int jbd2_journal_create_slab(size_t slab_size); 102 103 #ifdef CONFIG_JBD2_DEBUG 104 void __jbd2_debug(int level, const char *file, const char *func, 105 unsigned int line, const char *fmt, ...) 106 { 107 struct va_format vaf; 108 va_list args; 109 110 if (level > jbd2_journal_enable_debug) 111 return; 112 va_start(args, fmt); 113 vaf.fmt = fmt; 114 vaf.va = &args; 115 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf); 116 va_end(args); 117 } 118 EXPORT_SYMBOL(__jbd2_debug); 119 #endif 120 121 /* Checksumming functions */ 122 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) 123 { 124 if (!jbd2_journal_has_csum_v2or3_feature(j)) 125 return 1; 126 127 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; 128 } 129 130 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) 131 { 132 __u32 csum; 133 __be32 old_csum; 134 135 old_csum = sb->s_checksum; 136 sb->s_checksum = 0; 137 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); 138 sb->s_checksum = old_csum; 139 140 return cpu_to_be32(csum); 141 } 142 143 /* 144 * Helper function used to manage commit timeouts 145 */ 146 147 static void commit_timeout(struct timer_list *t) 148 { 149 journal_t *journal = from_timer(journal, t, j_commit_timer); 150 151 wake_up_process(journal->j_task); 152 } 153 154 /* 155 * kjournald2: The main thread function used to manage a logging device 156 * journal. 157 * 158 * This kernel thread is responsible for two things: 159 * 160 * 1) COMMIT: Every so often we need to commit the current state of the 161 * filesystem to disk. The journal thread is responsible for writing 162 * all of the metadata buffers to disk. If a fast commit is ongoing 163 * journal thread waits until it's done and then continues from 164 * there on. 165 * 166 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all 167 * of the data in that part of the log has been rewritten elsewhere on 168 * the disk. Flushing these old buffers to reclaim space in the log is 169 * known as checkpointing, and this thread is responsible for that job. 170 */ 171 172 static int kjournald2(void *arg) 173 { 174 journal_t *journal = arg; 175 transaction_t *transaction; 176 177 /* 178 * Set up an interval timer which can be used to trigger a commit wakeup 179 * after the commit interval expires 180 */ 181 timer_setup(&journal->j_commit_timer, commit_timeout, 0); 182 183 set_freezable(); 184 185 /* Record that the journal thread is running */ 186 journal->j_task = current; 187 wake_up(&journal->j_wait_done_commit); 188 189 /* 190 * Make sure that no allocations from this kernel thread will ever 191 * recurse to the fs layer because we are responsible for the 192 * transaction commit and any fs involvement might get stuck waiting for 193 * the trasn. commit. 194 */ 195 memalloc_nofs_save(); 196 197 /* 198 * And now, wait forever for commit wakeup events. 199 */ 200 write_lock(&journal->j_state_lock); 201 202 loop: 203 if (journal->j_flags & JBD2_UNMOUNT) 204 goto end_loop; 205 206 jbd_debug(1, "commit_sequence=%u, commit_request=%u\n", 207 journal->j_commit_sequence, journal->j_commit_request); 208 209 if (journal->j_commit_sequence != journal->j_commit_request) { 210 jbd_debug(1, "OK, requests differ\n"); 211 write_unlock(&journal->j_state_lock); 212 del_timer_sync(&journal->j_commit_timer); 213 jbd2_journal_commit_transaction(journal); 214 write_lock(&journal->j_state_lock); 215 goto loop; 216 } 217 218 wake_up(&journal->j_wait_done_commit); 219 if (freezing(current)) { 220 /* 221 * The simpler the better. Flushing journal isn't a 222 * good idea, because that depends on threads that may 223 * be already stopped. 224 */ 225 jbd_debug(1, "Now suspending kjournald2\n"); 226 write_unlock(&journal->j_state_lock); 227 try_to_freeze(); 228 write_lock(&journal->j_state_lock); 229 } else { 230 /* 231 * We assume on resume that commits are already there, 232 * so we don't sleep 233 */ 234 DEFINE_WAIT(wait); 235 int should_sleep = 1; 236 237 prepare_to_wait(&journal->j_wait_commit, &wait, 238 TASK_INTERRUPTIBLE); 239 if (journal->j_commit_sequence != journal->j_commit_request) 240 should_sleep = 0; 241 transaction = journal->j_running_transaction; 242 if (transaction && time_after_eq(jiffies, 243 transaction->t_expires)) 244 should_sleep = 0; 245 if (journal->j_flags & JBD2_UNMOUNT) 246 should_sleep = 0; 247 if (should_sleep) { 248 write_unlock(&journal->j_state_lock); 249 schedule(); 250 write_lock(&journal->j_state_lock); 251 } 252 finish_wait(&journal->j_wait_commit, &wait); 253 } 254 255 jbd_debug(1, "kjournald2 wakes\n"); 256 257 /* 258 * Were we woken up by a commit wakeup event? 259 */ 260 transaction = journal->j_running_transaction; 261 if (transaction && time_after_eq(jiffies, transaction->t_expires)) { 262 journal->j_commit_request = transaction->t_tid; 263 jbd_debug(1, "woke because of timeout\n"); 264 } 265 goto loop; 266 267 end_loop: 268 del_timer_sync(&journal->j_commit_timer); 269 journal->j_task = NULL; 270 wake_up(&journal->j_wait_done_commit); 271 jbd_debug(1, "Journal thread exiting.\n"); 272 write_unlock(&journal->j_state_lock); 273 return 0; 274 } 275 276 static int jbd2_journal_start_thread(journal_t *journal) 277 { 278 struct task_struct *t; 279 280 t = kthread_run(kjournald2, journal, "jbd2/%s", 281 journal->j_devname); 282 if (IS_ERR(t)) 283 return PTR_ERR(t); 284 285 wait_event(journal->j_wait_done_commit, journal->j_task != NULL); 286 return 0; 287 } 288 289 static void journal_kill_thread(journal_t *journal) 290 { 291 write_lock(&journal->j_state_lock); 292 journal->j_flags |= JBD2_UNMOUNT; 293 294 while (journal->j_task) { 295 write_unlock(&journal->j_state_lock); 296 wake_up(&journal->j_wait_commit); 297 wait_event(journal->j_wait_done_commit, journal->j_task == NULL); 298 write_lock(&journal->j_state_lock); 299 } 300 write_unlock(&journal->j_state_lock); 301 } 302 303 /* 304 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. 305 * 306 * Writes a metadata buffer to a given disk block. The actual IO is not 307 * performed but a new buffer_head is constructed which labels the data 308 * to be written with the correct destination disk block. 309 * 310 * Any magic-number escaping which needs to be done will cause a 311 * copy-out here. If the buffer happens to start with the 312 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the 313 * magic number is only written to the log for descripter blocks. In 314 * this case, we copy the data and replace the first word with 0, and we 315 * return a result code which indicates that this buffer needs to be 316 * marked as an escaped buffer in the corresponding log descriptor 317 * block. The missing word can then be restored when the block is read 318 * during recovery. 319 * 320 * If the source buffer has already been modified by a new transaction 321 * since we took the last commit snapshot, we use the frozen copy of 322 * that data for IO. If we end up using the existing buffer_head's data 323 * for the write, then we have to make sure nobody modifies it while the 324 * IO is in progress. do_get_write_access() handles this. 325 * 326 * The function returns a pointer to the buffer_head to be used for IO. 327 * 328 * 329 * Return value: 330 * <0: Error 331 * >=0: Finished OK 332 * 333 * On success: 334 * Bit 0 set == escape performed on the data 335 * Bit 1 set == buffer copy-out performed (kfree the data after IO) 336 */ 337 338 int jbd2_journal_write_metadata_buffer(transaction_t *transaction, 339 struct journal_head *jh_in, 340 struct buffer_head **bh_out, 341 sector_t blocknr) 342 { 343 int need_copy_out = 0; 344 int done_copy_out = 0; 345 int do_escape = 0; 346 char *mapped_data; 347 struct buffer_head *new_bh; 348 struct page *new_page; 349 unsigned int new_offset; 350 struct buffer_head *bh_in = jh2bh(jh_in); 351 journal_t *journal = transaction->t_journal; 352 353 /* 354 * The buffer really shouldn't be locked: only the current committing 355 * transaction is allowed to write it, so nobody else is allowed 356 * to do any IO. 357 * 358 * akpm: except if we're journalling data, and write() output is 359 * also part of a shared mapping, and another thread has 360 * decided to launch a writepage() against this buffer. 361 */ 362 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); 363 364 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); 365 366 /* keep subsequent assertions sane */ 367 atomic_set(&new_bh->b_count, 1); 368 369 spin_lock(&jh_in->b_state_lock); 370 repeat: 371 /* 372 * If a new transaction has already done a buffer copy-out, then 373 * we use that version of the data for the commit. 374 */ 375 if (jh_in->b_frozen_data) { 376 done_copy_out = 1; 377 new_page = virt_to_page(jh_in->b_frozen_data); 378 new_offset = offset_in_page(jh_in->b_frozen_data); 379 } else { 380 new_page = jh2bh(jh_in)->b_page; 381 new_offset = offset_in_page(jh2bh(jh_in)->b_data); 382 } 383 384 mapped_data = kmap_atomic(new_page); 385 /* 386 * Fire data frozen trigger if data already wasn't frozen. Do this 387 * before checking for escaping, as the trigger may modify the magic 388 * offset. If a copy-out happens afterwards, it will have the correct 389 * data in the buffer. 390 */ 391 if (!done_copy_out) 392 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, 393 jh_in->b_triggers); 394 395 /* 396 * Check for escaping 397 */ 398 if (*((__be32 *)(mapped_data + new_offset)) == 399 cpu_to_be32(JBD2_MAGIC_NUMBER)) { 400 need_copy_out = 1; 401 do_escape = 1; 402 } 403 kunmap_atomic(mapped_data); 404 405 /* 406 * Do we need to do a data copy? 407 */ 408 if (need_copy_out && !done_copy_out) { 409 char *tmp; 410 411 spin_unlock(&jh_in->b_state_lock); 412 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); 413 if (!tmp) { 414 brelse(new_bh); 415 return -ENOMEM; 416 } 417 spin_lock(&jh_in->b_state_lock); 418 if (jh_in->b_frozen_data) { 419 jbd2_free(tmp, bh_in->b_size); 420 goto repeat; 421 } 422 423 jh_in->b_frozen_data = tmp; 424 mapped_data = kmap_atomic(new_page); 425 memcpy(tmp, mapped_data + new_offset, bh_in->b_size); 426 kunmap_atomic(mapped_data); 427 428 new_page = virt_to_page(tmp); 429 new_offset = offset_in_page(tmp); 430 done_copy_out = 1; 431 432 /* 433 * This isn't strictly necessary, as we're using frozen 434 * data for the escaping, but it keeps consistency with 435 * b_frozen_data usage. 436 */ 437 jh_in->b_frozen_triggers = jh_in->b_triggers; 438 } 439 440 /* 441 * Did we need to do an escaping? Now we've done all the 442 * copying, we can finally do so. 443 */ 444 if (do_escape) { 445 mapped_data = kmap_atomic(new_page); 446 *((unsigned int *)(mapped_data + new_offset)) = 0; 447 kunmap_atomic(mapped_data); 448 } 449 450 set_bh_page(new_bh, new_page, new_offset); 451 new_bh->b_size = bh_in->b_size; 452 new_bh->b_bdev = journal->j_dev; 453 new_bh->b_blocknr = blocknr; 454 new_bh->b_private = bh_in; 455 set_buffer_mapped(new_bh); 456 set_buffer_dirty(new_bh); 457 458 *bh_out = new_bh; 459 460 /* 461 * The to-be-written buffer needs to get moved to the io queue, 462 * and the original buffer whose contents we are shadowing or 463 * copying is moved to the transaction's shadow queue. 464 */ 465 JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); 466 spin_lock(&journal->j_list_lock); 467 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); 468 spin_unlock(&journal->j_list_lock); 469 set_buffer_shadow(bh_in); 470 spin_unlock(&jh_in->b_state_lock); 471 472 return do_escape | (done_copy_out << 1); 473 } 474 475 /* 476 * Allocation code for the journal file. Manage the space left in the 477 * journal, so that we can begin checkpointing when appropriate. 478 */ 479 480 /* 481 * Called with j_state_lock locked for writing. 482 * Returns true if a transaction commit was started. 483 */ 484 int __jbd2_log_start_commit(journal_t *journal, tid_t target) 485 { 486 /* Return if the txn has already requested to be committed */ 487 if (journal->j_commit_request == target) 488 return 0; 489 490 /* 491 * The only transaction we can possibly wait upon is the 492 * currently running transaction (if it exists). Otherwise, 493 * the target tid must be an old one. 494 */ 495 if (journal->j_running_transaction && 496 journal->j_running_transaction->t_tid == target) { 497 /* 498 * We want a new commit: OK, mark the request and wakeup the 499 * commit thread. We do _not_ do the commit ourselves. 500 */ 501 502 journal->j_commit_request = target; 503 jbd_debug(1, "JBD2: requesting commit %u/%u\n", 504 journal->j_commit_request, 505 journal->j_commit_sequence); 506 journal->j_running_transaction->t_requested = jiffies; 507 wake_up(&journal->j_wait_commit); 508 return 1; 509 } else if (!tid_geq(journal->j_commit_request, target)) 510 /* This should never happen, but if it does, preserve 511 the evidence before kjournald goes into a loop and 512 increments j_commit_sequence beyond all recognition. */ 513 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", 514 journal->j_commit_request, 515 journal->j_commit_sequence, 516 target, journal->j_running_transaction ? 517 journal->j_running_transaction->t_tid : 0); 518 return 0; 519 } 520 521 int jbd2_log_start_commit(journal_t *journal, tid_t tid) 522 { 523 int ret; 524 525 write_lock(&journal->j_state_lock); 526 ret = __jbd2_log_start_commit(journal, tid); 527 write_unlock(&journal->j_state_lock); 528 return ret; 529 } 530 531 /* 532 * Force and wait any uncommitted transactions. We can only force the running 533 * transaction if we don't have an active handle, otherwise, we will deadlock. 534 * Returns: <0 in case of error, 535 * 0 if nothing to commit, 536 * 1 if transaction was successfully committed. 537 */ 538 static int __jbd2_journal_force_commit(journal_t *journal) 539 { 540 transaction_t *transaction = NULL; 541 tid_t tid; 542 int need_to_start = 0, ret = 0; 543 544 read_lock(&journal->j_state_lock); 545 if (journal->j_running_transaction && !current->journal_info) { 546 transaction = journal->j_running_transaction; 547 if (!tid_geq(journal->j_commit_request, transaction->t_tid)) 548 need_to_start = 1; 549 } else if (journal->j_committing_transaction) 550 transaction = journal->j_committing_transaction; 551 552 if (!transaction) { 553 /* Nothing to commit */ 554 read_unlock(&journal->j_state_lock); 555 return 0; 556 } 557 tid = transaction->t_tid; 558 read_unlock(&journal->j_state_lock); 559 if (need_to_start) 560 jbd2_log_start_commit(journal, tid); 561 ret = jbd2_log_wait_commit(journal, tid); 562 if (!ret) 563 ret = 1; 564 565 return ret; 566 } 567 568 /** 569 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the 570 * calling process is not within transaction. 571 * 572 * @journal: journal to force 573 * Returns true if progress was made. 574 * 575 * This is used for forcing out undo-protected data which contains 576 * bitmaps, when the fs is running out of space. 577 */ 578 int jbd2_journal_force_commit_nested(journal_t *journal) 579 { 580 int ret; 581 582 ret = __jbd2_journal_force_commit(journal); 583 return ret > 0; 584 } 585 586 /** 587 * jbd2_journal_force_commit() - force any uncommitted transactions 588 * @journal: journal to force 589 * 590 * Caller want unconditional commit. We can only force the running transaction 591 * if we don't have an active handle, otherwise, we will deadlock. 592 */ 593 int jbd2_journal_force_commit(journal_t *journal) 594 { 595 int ret; 596 597 J_ASSERT(!current->journal_info); 598 ret = __jbd2_journal_force_commit(journal); 599 if (ret > 0) 600 ret = 0; 601 return ret; 602 } 603 604 /* 605 * Start a commit of the current running transaction (if any). Returns true 606 * if a transaction is going to be committed (or is currently already 607 * committing), and fills its tid in at *ptid 608 */ 609 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) 610 { 611 int ret = 0; 612 613 write_lock(&journal->j_state_lock); 614 if (journal->j_running_transaction) { 615 tid_t tid = journal->j_running_transaction->t_tid; 616 617 __jbd2_log_start_commit(journal, tid); 618 /* There's a running transaction and we've just made sure 619 * it's commit has been scheduled. */ 620 if (ptid) 621 *ptid = tid; 622 ret = 1; 623 } else if (journal->j_committing_transaction) { 624 /* 625 * If commit has been started, then we have to wait for 626 * completion of that transaction. 627 */ 628 if (ptid) 629 *ptid = journal->j_committing_transaction->t_tid; 630 ret = 1; 631 } 632 write_unlock(&journal->j_state_lock); 633 return ret; 634 } 635 636 /* 637 * Return 1 if a given transaction has not yet sent barrier request 638 * connected with a transaction commit. If 0 is returned, transaction 639 * may or may not have sent the barrier. Used to avoid sending barrier 640 * twice in common cases. 641 */ 642 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) 643 { 644 int ret = 0; 645 transaction_t *commit_trans; 646 647 if (!(journal->j_flags & JBD2_BARRIER)) 648 return 0; 649 read_lock(&journal->j_state_lock); 650 /* Transaction already committed? */ 651 if (tid_geq(journal->j_commit_sequence, tid)) 652 goto out; 653 commit_trans = journal->j_committing_transaction; 654 if (!commit_trans || commit_trans->t_tid != tid) { 655 ret = 1; 656 goto out; 657 } 658 /* 659 * Transaction is being committed and we already proceeded to 660 * submitting a flush to fs partition? 661 */ 662 if (journal->j_fs_dev != journal->j_dev) { 663 if (!commit_trans->t_need_data_flush || 664 commit_trans->t_state >= T_COMMIT_DFLUSH) 665 goto out; 666 } else { 667 if (commit_trans->t_state >= T_COMMIT_JFLUSH) 668 goto out; 669 } 670 ret = 1; 671 out: 672 read_unlock(&journal->j_state_lock); 673 return ret; 674 } 675 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); 676 677 /* 678 * Wait for a specified commit to complete. 679 * The caller may not hold the journal lock. 680 */ 681 int jbd2_log_wait_commit(journal_t *journal, tid_t tid) 682 { 683 int err = 0; 684 685 read_lock(&journal->j_state_lock); 686 #ifdef CONFIG_PROVE_LOCKING 687 /* 688 * Some callers make sure transaction is already committing and in that 689 * case we cannot block on open handles anymore. So don't warn in that 690 * case. 691 */ 692 if (tid_gt(tid, journal->j_commit_sequence) && 693 (!journal->j_committing_transaction || 694 journal->j_committing_transaction->t_tid != tid)) { 695 read_unlock(&journal->j_state_lock); 696 jbd2_might_wait_for_commit(journal); 697 read_lock(&journal->j_state_lock); 698 } 699 #endif 700 #ifdef CONFIG_JBD2_DEBUG 701 if (!tid_geq(journal->j_commit_request, tid)) { 702 printk(KERN_ERR 703 "%s: error: j_commit_request=%u, tid=%u\n", 704 __func__, journal->j_commit_request, tid); 705 } 706 #endif 707 while (tid_gt(tid, journal->j_commit_sequence)) { 708 jbd_debug(1, "JBD2: want %u, j_commit_sequence=%u\n", 709 tid, journal->j_commit_sequence); 710 read_unlock(&journal->j_state_lock); 711 wake_up(&journal->j_wait_commit); 712 wait_event(journal->j_wait_done_commit, 713 !tid_gt(tid, journal->j_commit_sequence)); 714 read_lock(&journal->j_state_lock); 715 } 716 read_unlock(&journal->j_state_lock); 717 718 if (unlikely(is_journal_aborted(journal))) 719 err = -EIO; 720 return err; 721 } 722 723 /* 724 * Start a fast commit. If there's an ongoing fast or full commit wait for 725 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY 726 * if a fast commit is not needed, either because there's an already a commit 727 * going on or this tid has already been committed. Returns -EINVAL if no jbd2 728 * commit has yet been performed. 729 */ 730 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid) 731 { 732 if (unlikely(is_journal_aborted(journal))) 733 return -EIO; 734 /* 735 * Fast commits only allowed if at least one full commit has 736 * been processed. 737 */ 738 if (!journal->j_stats.ts_tid) 739 return -EINVAL; 740 741 write_lock(&journal->j_state_lock); 742 if (tid <= journal->j_commit_sequence) { 743 write_unlock(&journal->j_state_lock); 744 return -EALREADY; 745 } 746 747 if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING || 748 (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) { 749 DEFINE_WAIT(wait); 750 751 prepare_to_wait(&journal->j_fc_wait, &wait, 752 TASK_UNINTERRUPTIBLE); 753 write_unlock(&journal->j_state_lock); 754 schedule(); 755 finish_wait(&journal->j_fc_wait, &wait); 756 return -EALREADY; 757 } 758 journal->j_flags |= JBD2_FAST_COMMIT_ONGOING; 759 write_unlock(&journal->j_state_lock); 760 jbd2_journal_lock_updates(journal); 761 762 return 0; 763 } 764 EXPORT_SYMBOL(jbd2_fc_begin_commit); 765 766 /* 767 * Stop a fast commit. If fallback is set, this function starts commit of 768 * TID tid before any other fast commit can start. 769 */ 770 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback) 771 { 772 jbd2_journal_unlock_updates(journal); 773 if (journal->j_fc_cleanup_callback) 774 journal->j_fc_cleanup_callback(journal, 0, tid); 775 write_lock(&journal->j_state_lock); 776 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING; 777 if (fallback) 778 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING; 779 write_unlock(&journal->j_state_lock); 780 wake_up(&journal->j_fc_wait); 781 if (fallback) 782 return jbd2_complete_transaction(journal, tid); 783 return 0; 784 } 785 786 int jbd2_fc_end_commit(journal_t *journal) 787 { 788 return __jbd2_fc_end_commit(journal, 0, false); 789 } 790 EXPORT_SYMBOL(jbd2_fc_end_commit); 791 792 int jbd2_fc_end_commit_fallback(journal_t *journal) 793 { 794 tid_t tid; 795 796 read_lock(&journal->j_state_lock); 797 tid = journal->j_running_transaction ? 798 journal->j_running_transaction->t_tid : 0; 799 read_unlock(&journal->j_state_lock); 800 return __jbd2_fc_end_commit(journal, tid, true); 801 } 802 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback); 803 804 /* Return 1 when transaction with given tid has already committed. */ 805 int jbd2_transaction_committed(journal_t *journal, tid_t tid) 806 { 807 int ret = 1; 808 809 read_lock(&journal->j_state_lock); 810 if (journal->j_running_transaction && 811 journal->j_running_transaction->t_tid == tid) 812 ret = 0; 813 if (journal->j_committing_transaction && 814 journal->j_committing_transaction->t_tid == tid) 815 ret = 0; 816 read_unlock(&journal->j_state_lock); 817 return ret; 818 } 819 EXPORT_SYMBOL(jbd2_transaction_committed); 820 821 /* 822 * When this function returns the transaction corresponding to tid 823 * will be completed. If the transaction has currently running, start 824 * committing that transaction before waiting for it to complete. If 825 * the transaction id is stale, it is by definition already completed, 826 * so just return SUCCESS. 827 */ 828 int jbd2_complete_transaction(journal_t *journal, tid_t tid) 829 { 830 int need_to_wait = 1; 831 832 read_lock(&journal->j_state_lock); 833 if (journal->j_running_transaction && 834 journal->j_running_transaction->t_tid == tid) { 835 if (journal->j_commit_request != tid) { 836 /* transaction not yet started, so request it */ 837 read_unlock(&journal->j_state_lock); 838 jbd2_log_start_commit(journal, tid); 839 goto wait_commit; 840 } 841 } else if (!(journal->j_committing_transaction && 842 journal->j_committing_transaction->t_tid == tid)) 843 need_to_wait = 0; 844 read_unlock(&journal->j_state_lock); 845 if (!need_to_wait) 846 return 0; 847 wait_commit: 848 return jbd2_log_wait_commit(journal, tid); 849 } 850 EXPORT_SYMBOL(jbd2_complete_transaction); 851 852 /* 853 * Log buffer allocation routines: 854 */ 855 856 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) 857 { 858 unsigned long blocknr; 859 860 write_lock(&journal->j_state_lock); 861 J_ASSERT(journal->j_free > 1); 862 863 blocknr = journal->j_head; 864 journal->j_head++; 865 journal->j_free--; 866 if (journal->j_head == journal->j_last) 867 journal->j_head = journal->j_first; 868 write_unlock(&journal->j_state_lock); 869 return jbd2_journal_bmap(journal, blocknr, retp); 870 } 871 872 /* Map one fast commit buffer for use by the file system */ 873 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out) 874 { 875 unsigned long long pblock; 876 unsigned long blocknr; 877 int ret = 0; 878 struct buffer_head *bh; 879 int fc_off; 880 881 *bh_out = NULL; 882 883 if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) { 884 fc_off = journal->j_fc_off; 885 blocknr = journal->j_fc_first + fc_off; 886 journal->j_fc_off++; 887 } else { 888 ret = -EINVAL; 889 } 890 891 if (ret) 892 return ret; 893 894 ret = jbd2_journal_bmap(journal, blocknr, &pblock); 895 if (ret) 896 return ret; 897 898 bh = __getblk(journal->j_dev, pblock, journal->j_blocksize); 899 if (!bh) 900 return -ENOMEM; 901 902 903 journal->j_fc_wbuf[fc_off] = bh; 904 905 *bh_out = bh; 906 907 return 0; 908 } 909 EXPORT_SYMBOL(jbd2_fc_get_buf); 910 911 /* 912 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf 913 * for completion. 914 */ 915 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks) 916 { 917 struct buffer_head *bh; 918 int i, j_fc_off; 919 920 j_fc_off = journal->j_fc_off; 921 922 /* 923 * Wait in reverse order to minimize chances of us being woken up before 924 * all IOs have completed 925 */ 926 for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) { 927 bh = journal->j_fc_wbuf[i]; 928 wait_on_buffer(bh); 929 put_bh(bh); 930 journal->j_fc_wbuf[i] = NULL; 931 if (unlikely(!buffer_uptodate(bh))) 932 return -EIO; 933 } 934 935 return 0; 936 } 937 EXPORT_SYMBOL(jbd2_fc_wait_bufs); 938 939 int jbd2_fc_release_bufs(journal_t *journal) 940 { 941 struct buffer_head *bh; 942 int i, j_fc_off; 943 944 j_fc_off = journal->j_fc_off; 945 946 for (i = j_fc_off - 1; i >= 0; i--) { 947 bh = journal->j_fc_wbuf[i]; 948 if (!bh) 949 break; 950 put_bh(bh); 951 journal->j_fc_wbuf[i] = NULL; 952 } 953 954 return 0; 955 } 956 EXPORT_SYMBOL(jbd2_fc_release_bufs); 957 958 /* 959 * Conversion of logical to physical block numbers for the journal 960 * 961 * On external journals the journal blocks are identity-mapped, so 962 * this is a no-op. If needed, we can use j_blk_offset - everything is 963 * ready. 964 */ 965 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, 966 unsigned long long *retp) 967 { 968 int err = 0; 969 unsigned long long ret; 970 sector_t block = 0; 971 972 if (journal->j_inode) { 973 block = blocknr; 974 ret = bmap(journal->j_inode, &block); 975 976 if (ret || !block) { 977 printk(KERN_ALERT "%s: journal block not found " 978 "at offset %lu on %s\n", 979 __func__, blocknr, journal->j_devname); 980 err = -EIO; 981 jbd2_journal_abort(journal, err); 982 } else { 983 *retp = block; 984 } 985 986 } else { 987 *retp = blocknr; /* +journal->j_blk_offset */ 988 } 989 return err; 990 } 991 992 /* 993 * We play buffer_head aliasing tricks to write data/metadata blocks to 994 * the journal without copying their contents, but for journal 995 * descriptor blocks we do need to generate bona fide buffers. 996 * 997 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying 998 * the buffer's contents they really should run flush_dcache_page(bh->b_page). 999 * But we don't bother doing that, so there will be coherency problems with 1000 * mmaps of blockdevs which hold live JBD-controlled filesystems. 1001 */ 1002 struct buffer_head * 1003 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type) 1004 { 1005 journal_t *journal = transaction->t_journal; 1006 struct buffer_head *bh; 1007 unsigned long long blocknr; 1008 journal_header_t *header; 1009 int err; 1010 1011 err = jbd2_journal_next_log_block(journal, &blocknr); 1012 1013 if (err) 1014 return NULL; 1015 1016 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); 1017 if (!bh) 1018 return NULL; 1019 atomic_dec(&transaction->t_outstanding_credits); 1020 lock_buffer(bh); 1021 memset(bh->b_data, 0, journal->j_blocksize); 1022 header = (journal_header_t *)bh->b_data; 1023 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); 1024 header->h_blocktype = cpu_to_be32(type); 1025 header->h_sequence = cpu_to_be32(transaction->t_tid); 1026 set_buffer_uptodate(bh); 1027 unlock_buffer(bh); 1028 BUFFER_TRACE(bh, "return this buffer"); 1029 return bh; 1030 } 1031 1032 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh) 1033 { 1034 struct jbd2_journal_block_tail *tail; 1035 __u32 csum; 1036 1037 if (!jbd2_journal_has_csum_v2or3(j)) 1038 return; 1039 1040 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - 1041 sizeof(struct jbd2_journal_block_tail)); 1042 tail->t_checksum = 0; 1043 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); 1044 tail->t_checksum = cpu_to_be32(csum); 1045 } 1046 1047 /* 1048 * Return tid of the oldest transaction in the journal and block in the journal 1049 * where the transaction starts. 1050 * 1051 * If the journal is now empty, return which will be the next transaction ID 1052 * we will write and where will that transaction start. 1053 * 1054 * The return value is 0 if journal tail cannot be pushed any further, 1 if 1055 * it can. 1056 */ 1057 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, 1058 unsigned long *block) 1059 { 1060 transaction_t *transaction; 1061 int ret; 1062 1063 read_lock(&journal->j_state_lock); 1064 spin_lock(&journal->j_list_lock); 1065 transaction = journal->j_checkpoint_transactions; 1066 if (transaction) { 1067 *tid = transaction->t_tid; 1068 *block = transaction->t_log_start; 1069 } else if ((transaction = journal->j_committing_transaction) != NULL) { 1070 *tid = transaction->t_tid; 1071 *block = transaction->t_log_start; 1072 } else if ((transaction = journal->j_running_transaction) != NULL) { 1073 *tid = transaction->t_tid; 1074 *block = journal->j_head; 1075 } else { 1076 *tid = journal->j_transaction_sequence; 1077 *block = journal->j_head; 1078 } 1079 ret = tid_gt(*tid, journal->j_tail_sequence); 1080 spin_unlock(&journal->j_list_lock); 1081 read_unlock(&journal->j_state_lock); 1082 1083 return ret; 1084 } 1085 1086 /* 1087 * Update information in journal structure and in on disk journal superblock 1088 * about log tail. This function does not check whether information passed in 1089 * really pushes log tail further. It's responsibility of the caller to make 1090 * sure provided log tail information is valid (e.g. by holding 1091 * j_checkpoint_mutex all the time between computing log tail and calling this 1092 * function as is the case with jbd2_cleanup_journal_tail()). 1093 * 1094 * Requires j_checkpoint_mutex 1095 */ 1096 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1097 { 1098 unsigned long freed; 1099 int ret; 1100 1101 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1102 1103 /* 1104 * We cannot afford for write to remain in drive's caches since as 1105 * soon as we update j_tail, next transaction can start reusing journal 1106 * space and if we lose sb update during power failure we'd replay 1107 * old transaction with possibly newly overwritten data. 1108 */ 1109 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, 1110 REQ_SYNC | REQ_FUA); 1111 if (ret) 1112 goto out; 1113 1114 write_lock(&journal->j_state_lock); 1115 freed = block - journal->j_tail; 1116 if (block < journal->j_tail) 1117 freed += journal->j_last - journal->j_first; 1118 1119 trace_jbd2_update_log_tail(journal, tid, block, freed); 1120 jbd_debug(1, 1121 "Cleaning journal tail from %u to %u (offset %lu), " 1122 "freeing %lu\n", 1123 journal->j_tail_sequence, tid, block, freed); 1124 1125 journal->j_free += freed; 1126 journal->j_tail_sequence = tid; 1127 journal->j_tail = block; 1128 write_unlock(&journal->j_state_lock); 1129 1130 out: 1131 return ret; 1132 } 1133 1134 /* 1135 * This is a variation of __jbd2_update_log_tail which checks for validity of 1136 * provided log tail and locks j_checkpoint_mutex. So it is safe against races 1137 * with other threads updating log tail. 1138 */ 1139 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1140 { 1141 mutex_lock_io(&journal->j_checkpoint_mutex); 1142 if (tid_gt(tid, journal->j_tail_sequence)) 1143 __jbd2_update_log_tail(journal, tid, block); 1144 mutex_unlock(&journal->j_checkpoint_mutex); 1145 } 1146 1147 struct jbd2_stats_proc_session { 1148 journal_t *journal; 1149 struct transaction_stats_s *stats; 1150 int start; 1151 int max; 1152 }; 1153 1154 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) 1155 { 1156 return *pos ? NULL : SEQ_START_TOKEN; 1157 } 1158 1159 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) 1160 { 1161 (*pos)++; 1162 return NULL; 1163 } 1164 1165 static int jbd2_seq_info_show(struct seq_file *seq, void *v) 1166 { 1167 struct jbd2_stats_proc_session *s = seq->private; 1168 1169 if (v != SEQ_START_TOKEN) 1170 return 0; 1171 seq_printf(seq, "%lu transactions (%lu requested), " 1172 "each up to %u blocks\n", 1173 s->stats->ts_tid, s->stats->ts_requested, 1174 s->journal->j_max_transaction_buffers); 1175 if (s->stats->ts_tid == 0) 1176 return 0; 1177 seq_printf(seq, "average: \n %ums waiting for transaction\n", 1178 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); 1179 seq_printf(seq, " %ums request delay\n", 1180 (s->stats->ts_requested == 0) ? 0 : 1181 jiffies_to_msecs(s->stats->run.rs_request_delay / 1182 s->stats->ts_requested)); 1183 seq_printf(seq, " %ums running transaction\n", 1184 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); 1185 seq_printf(seq, " %ums transaction was being locked\n", 1186 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); 1187 seq_printf(seq, " %ums flushing data (in ordered mode)\n", 1188 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); 1189 seq_printf(seq, " %ums logging transaction\n", 1190 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); 1191 seq_printf(seq, " %lluus average transaction commit time\n", 1192 div_u64(s->journal->j_average_commit_time, 1000)); 1193 seq_printf(seq, " %lu handles per transaction\n", 1194 s->stats->run.rs_handle_count / s->stats->ts_tid); 1195 seq_printf(seq, " %lu blocks per transaction\n", 1196 s->stats->run.rs_blocks / s->stats->ts_tid); 1197 seq_printf(seq, " %lu logged blocks per transaction\n", 1198 s->stats->run.rs_blocks_logged / s->stats->ts_tid); 1199 return 0; 1200 } 1201 1202 static void jbd2_seq_info_stop(struct seq_file *seq, void *v) 1203 { 1204 } 1205 1206 static const struct seq_operations jbd2_seq_info_ops = { 1207 .start = jbd2_seq_info_start, 1208 .next = jbd2_seq_info_next, 1209 .stop = jbd2_seq_info_stop, 1210 .show = jbd2_seq_info_show, 1211 }; 1212 1213 static int jbd2_seq_info_open(struct inode *inode, struct file *file) 1214 { 1215 journal_t *journal = pde_data(inode); 1216 struct jbd2_stats_proc_session *s; 1217 int rc, size; 1218 1219 s = kmalloc(sizeof(*s), GFP_KERNEL); 1220 if (s == NULL) 1221 return -ENOMEM; 1222 size = sizeof(struct transaction_stats_s); 1223 s->stats = kmalloc(size, GFP_KERNEL); 1224 if (s->stats == NULL) { 1225 kfree(s); 1226 return -ENOMEM; 1227 } 1228 spin_lock(&journal->j_history_lock); 1229 memcpy(s->stats, &journal->j_stats, size); 1230 s->journal = journal; 1231 spin_unlock(&journal->j_history_lock); 1232 1233 rc = seq_open(file, &jbd2_seq_info_ops); 1234 if (rc == 0) { 1235 struct seq_file *m = file->private_data; 1236 m->private = s; 1237 } else { 1238 kfree(s->stats); 1239 kfree(s); 1240 } 1241 return rc; 1242 1243 } 1244 1245 static int jbd2_seq_info_release(struct inode *inode, struct file *file) 1246 { 1247 struct seq_file *seq = file->private_data; 1248 struct jbd2_stats_proc_session *s = seq->private; 1249 kfree(s->stats); 1250 kfree(s); 1251 return seq_release(inode, file); 1252 } 1253 1254 static const struct proc_ops jbd2_info_proc_ops = { 1255 .proc_open = jbd2_seq_info_open, 1256 .proc_read = seq_read, 1257 .proc_lseek = seq_lseek, 1258 .proc_release = jbd2_seq_info_release, 1259 }; 1260 1261 static struct proc_dir_entry *proc_jbd2_stats; 1262 1263 static void jbd2_stats_proc_init(journal_t *journal) 1264 { 1265 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); 1266 if (journal->j_proc_entry) { 1267 proc_create_data("info", S_IRUGO, journal->j_proc_entry, 1268 &jbd2_info_proc_ops, journal); 1269 } 1270 } 1271 1272 static void jbd2_stats_proc_exit(journal_t *journal) 1273 { 1274 remove_proc_entry("info", journal->j_proc_entry); 1275 remove_proc_entry(journal->j_devname, proc_jbd2_stats); 1276 } 1277 1278 /* Minimum size of descriptor tag */ 1279 static int jbd2_min_tag_size(void) 1280 { 1281 /* 1282 * Tag with 32-bit block numbers does not use last four bytes of the 1283 * structure 1284 */ 1285 return sizeof(journal_block_tag_t) - 4; 1286 } 1287 1288 /** 1289 * jbd2_journal_shrink_scan() 1290 * @shrink: shrinker to work on 1291 * @sc: reclaim request to process 1292 * 1293 * Scan the checkpointed buffer on the checkpoint list and release the 1294 * journal_head. 1295 */ 1296 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink, 1297 struct shrink_control *sc) 1298 { 1299 journal_t *journal = container_of(shrink, journal_t, j_shrinker); 1300 unsigned long nr_to_scan = sc->nr_to_scan; 1301 unsigned long nr_shrunk; 1302 unsigned long count; 1303 1304 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1305 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count); 1306 1307 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan); 1308 1309 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1310 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count); 1311 1312 return nr_shrunk; 1313 } 1314 1315 /** 1316 * jbd2_journal_shrink_count() 1317 * @shrink: shrinker to work on 1318 * @sc: reclaim request to process 1319 * 1320 * Count the number of checkpoint buffers on the checkpoint list. 1321 */ 1322 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink, 1323 struct shrink_control *sc) 1324 { 1325 journal_t *journal = container_of(shrink, journal_t, j_shrinker); 1326 unsigned long count; 1327 1328 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1329 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count); 1330 1331 return count; 1332 } 1333 1334 /* 1335 * Management for journal control blocks: functions to create and 1336 * destroy journal_t structures, and to initialise and read existing 1337 * journal blocks from disk. */ 1338 1339 /* First: create and setup a journal_t object in memory. We initialise 1340 * very few fields yet: that has to wait until we have created the 1341 * journal structures from from scratch, or loaded them from disk. */ 1342 1343 static journal_t *journal_init_common(struct block_device *bdev, 1344 struct block_device *fs_dev, 1345 unsigned long long start, int len, int blocksize) 1346 { 1347 static struct lock_class_key jbd2_trans_commit_key; 1348 journal_t *journal; 1349 int err; 1350 struct buffer_head *bh; 1351 int n; 1352 1353 journal = kzalloc(sizeof(*journal), GFP_KERNEL); 1354 if (!journal) 1355 return NULL; 1356 1357 init_waitqueue_head(&journal->j_wait_transaction_locked); 1358 init_waitqueue_head(&journal->j_wait_done_commit); 1359 init_waitqueue_head(&journal->j_wait_commit); 1360 init_waitqueue_head(&journal->j_wait_updates); 1361 init_waitqueue_head(&journal->j_wait_reserved); 1362 init_waitqueue_head(&journal->j_fc_wait); 1363 mutex_init(&journal->j_abort_mutex); 1364 mutex_init(&journal->j_barrier); 1365 mutex_init(&journal->j_checkpoint_mutex); 1366 spin_lock_init(&journal->j_revoke_lock); 1367 spin_lock_init(&journal->j_list_lock); 1368 rwlock_init(&journal->j_state_lock); 1369 1370 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); 1371 journal->j_min_batch_time = 0; 1372 journal->j_max_batch_time = 15000; /* 15ms */ 1373 atomic_set(&journal->j_reserved_credits, 0); 1374 1375 /* The journal is marked for error until we succeed with recovery! */ 1376 journal->j_flags = JBD2_ABORT; 1377 1378 /* Set up a default-sized revoke table for the new mount. */ 1379 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); 1380 if (err) 1381 goto err_cleanup; 1382 1383 spin_lock_init(&journal->j_history_lock); 1384 1385 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle", 1386 &jbd2_trans_commit_key, 0); 1387 1388 /* journal descriptor can store up to n blocks -bzzz */ 1389 journal->j_blocksize = blocksize; 1390 journal->j_dev = bdev; 1391 journal->j_fs_dev = fs_dev; 1392 journal->j_blk_offset = start; 1393 journal->j_total_len = len; 1394 /* We need enough buffers to write out full descriptor block. */ 1395 n = journal->j_blocksize / jbd2_min_tag_size(); 1396 journal->j_wbufsize = n; 1397 journal->j_fc_wbuf = NULL; 1398 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *), 1399 GFP_KERNEL); 1400 if (!journal->j_wbuf) 1401 goto err_cleanup; 1402 1403 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize); 1404 if (!bh) { 1405 pr_err("%s: Cannot get buffer for journal superblock\n", 1406 __func__); 1407 goto err_cleanup; 1408 } 1409 journal->j_sb_buffer = bh; 1410 journal->j_superblock = (journal_superblock_t *)bh->b_data; 1411 1412 journal->j_shrink_transaction = NULL; 1413 journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan; 1414 journal->j_shrinker.count_objects = jbd2_journal_shrink_count; 1415 journal->j_shrinker.seeks = DEFAULT_SEEKS; 1416 journal->j_shrinker.batch = journal->j_max_transaction_buffers; 1417 1418 if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL)) 1419 goto err_cleanup; 1420 1421 if (register_shrinker(&journal->j_shrinker)) { 1422 percpu_counter_destroy(&journal->j_checkpoint_jh_count); 1423 goto err_cleanup; 1424 } 1425 return journal; 1426 1427 err_cleanup: 1428 brelse(journal->j_sb_buffer); 1429 kfree(journal->j_wbuf); 1430 jbd2_journal_destroy_revoke(journal); 1431 kfree(journal); 1432 return NULL; 1433 } 1434 1435 /* jbd2_journal_init_dev and jbd2_journal_init_inode: 1436 * 1437 * Create a journal structure assigned some fixed set of disk blocks to 1438 * the journal. We don't actually touch those disk blocks yet, but we 1439 * need to set up all of the mapping information to tell the journaling 1440 * system where the journal blocks are. 1441 * 1442 */ 1443 1444 /** 1445 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure 1446 * @bdev: Block device on which to create the journal 1447 * @fs_dev: Device which hold journalled filesystem for this journal. 1448 * @start: Block nr Start of journal. 1449 * @len: Length of the journal in blocks. 1450 * @blocksize: blocksize of journalling device 1451 * 1452 * Returns: a newly created journal_t * 1453 * 1454 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous 1455 * range of blocks on an arbitrary block device. 1456 * 1457 */ 1458 journal_t *jbd2_journal_init_dev(struct block_device *bdev, 1459 struct block_device *fs_dev, 1460 unsigned long long start, int len, int blocksize) 1461 { 1462 journal_t *journal; 1463 1464 journal = journal_init_common(bdev, fs_dev, start, len, blocksize); 1465 if (!journal) 1466 return NULL; 1467 1468 bdevname(journal->j_dev, journal->j_devname); 1469 strreplace(journal->j_devname, '/', '!'); 1470 jbd2_stats_proc_init(journal); 1471 1472 return journal; 1473 } 1474 1475 /** 1476 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. 1477 * @inode: An inode to create the journal in 1478 * 1479 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as 1480 * the journal. The inode must exist already, must support bmap() and 1481 * must have all data blocks preallocated. 1482 */ 1483 journal_t *jbd2_journal_init_inode(struct inode *inode) 1484 { 1485 journal_t *journal; 1486 sector_t blocknr; 1487 char *p; 1488 int err = 0; 1489 1490 blocknr = 0; 1491 err = bmap(inode, &blocknr); 1492 1493 if (err || !blocknr) { 1494 pr_err("%s: Cannot locate journal superblock\n", 1495 __func__); 1496 return NULL; 1497 } 1498 1499 jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n", 1500 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size, 1501 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); 1502 1503 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev, 1504 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits, 1505 inode->i_sb->s_blocksize); 1506 if (!journal) 1507 return NULL; 1508 1509 journal->j_inode = inode; 1510 bdevname(journal->j_dev, journal->j_devname); 1511 p = strreplace(journal->j_devname, '/', '!'); 1512 sprintf(p, "-%lu", journal->j_inode->i_ino); 1513 jbd2_stats_proc_init(journal); 1514 1515 return journal; 1516 } 1517 1518 /* 1519 * If the journal init or create aborts, we need to mark the journal 1520 * superblock as being NULL to prevent the journal destroy from writing 1521 * back a bogus superblock. 1522 */ 1523 static void journal_fail_superblock(journal_t *journal) 1524 { 1525 struct buffer_head *bh = journal->j_sb_buffer; 1526 brelse(bh); 1527 journal->j_sb_buffer = NULL; 1528 } 1529 1530 /* 1531 * Given a journal_t structure, initialise the various fields for 1532 * startup of a new journaling session. We use this both when creating 1533 * a journal, and after recovering an old journal to reset it for 1534 * subsequent use. 1535 */ 1536 1537 static int journal_reset(journal_t *journal) 1538 { 1539 journal_superblock_t *sb = journal->j_superblock; 1540 unsigned long long first, last; 1541 1542 first = be32_to_cpu(sb->s_first); 1543 last = be32_to_cpu(sb->s_maxlen); 1544 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { 1545 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", 1546 first, last); 1547 journal_fail_superblock(journal); 1548 return -EINVAL; 1549 } 1550 1551 journal->j_first = first; 1552 journal->j_last = last; 1553 1554 journal->j_head = journal->j_first; 1555 journal->j_tail = journal->j_first; 1556 journal->j_free = journal->j_last - journal->j_first; 1557 1558 journal->j_tail_sequence = journal->j_transaction_sequence; 1559 journal->j_commit_sequence = journal->j_transaction_sequence - 1; 1560 journal->j_commit_request = journal->j_commit_sequence; 1561 1562 journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal); 1563 1564 /* 1565 * Now that journal recovery is done, turn fast commits off here. This 1566 * way, if fast commit was enabled before the crash but if now FS has 1567 * disabled it, we don't enable fast commits. 1568 */ 1569 jbd2_clear_feature_fast_commit(journal); 1570 1571 /* 1572 * As a special case, if the on-disk copy is already marked as needing 1573 * no recovery (s_start == 0), then we can safely defer the superblock 1574 * update until the next commit by setting JBD2_FLUSHED. This avoids 1575 * attempting a write to a potential-readonly device. 1576 */ 1577 if (sb->s_start == 0) { 1578 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " 1579 "(start %ld, seq %u, errno %d)\n", 1580 journal->j_tail, journal->j_tail_sequence, 1581 journal->j_errno); 1582 journal->j_flags |= JBD2_FLUSHED; 1583 } else { 1584 /* Lock here to make assertions happy... */ 1585 mutex_lock_io(&journal->j_checkpoint_mutex); 1586 /* 1587 * Update log tail information. We use REQ_FUA since new 1588 * transaction will start reusing journal space and so we 1589 * must make sure information about current log tail is on 1590 * disk before that. 1591 */ 1592 jbd2_journal_update_sb_log_tail(journal, 1593 journal->j_tail_sequence, 1594 journal->j_tail, 1595 REQ_SYNC | REQ_FUA); 1596 mutex_unlock(&journal->j_checkpoint_mutex); 1597 } 1598 return jbd2_journal_start_thread(journal); 1599 } 1600 1601 /* 1602 * This function expects that the caller will have locked the journal 1603 * buffer head, and will return with it unlocked 1604 */ 1605 static int jbd2_write_superblock(journal_t *journal, int write_flags) 1606 { 1607 struct buffer_head *bh = journal->j_sb_buffer; 1608 journal_superblock_t *sb = journal->j_superblock; 1609 int ret; 1610 1611 /* Buffer got discarded which means block device got invalidated */ 1612 if (!buffer_mapped(bh)) { 1613 unlock_buffer(bh); 1614 return -EIO; 1615 } 1616 1617 trace_jbd2_write_superblock(journal, write_flags); 1618 if (!(journal->j_flags & JBD2_BARRIER)) 1619 write_flags &= ~(REQ_FUA | REQ_PREFLUSH); 1620 if (buffer_write_io_error(bh)) { 1621 /* 1622 * Oh, dear. A previous attempt to write the journal 1623 * superblock failed. This could happen because the 1624 * USB device was yanked out. Or it could happen to 1625 * be a transient write error and maybe the block will 1626 * be remapped. Nothing we can do but to retry the 1627 * write and hope for the best. 1628 */ 1629 printk(KERN_ERR "JBD2: previous I/O error detected " 1630 "for journal superblock update for %s.\n", 1631 journal->j_devname); 1632 clear_buffer_write_io_error(bh); 1633 set_buffer_uptodate(bh); 1634 } 1635 if (jbd2_journal_has_csum_v2or3(journal)) 1636 sb->s_checksum = jbd2_superblock_csum(journal, sb); 1637 get_bh(bh); 1638 bh->b_end_io = end_buffer_write_sync; 1639 ret = submit_bh(REQ_OP_WRITE, write_flags, bh); 1640 wait_on_buffer(bh); 1641 if (buffer_write_io_error(bh)) { 1642 clear_buffer_write_io_error(bh); 1643 set_buffer_uptodate(bh); 1644 ret = -EIO; 1645 } 1646 if (ret) { 1647 printk(KERN_ERR "JBD2: Error %d detected when updating " 1648 "journal superblock for %s.\n", ret, 1649 journal->j_devname); 1650 if (!is_journal_aborted(journal)) 1651 jbd2_journal_abort(journal, ret); 1652 } 1653 1654 return ret; 1655 } 1656 1657 /** 1658 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. 1659 * @journal: The journal to update. 1660 * @tail_tid: TID of the new transaction at the tail of the log 1661 * @tail_block: The first block of the transaction at the tail of the log 1662 * @write_op: With which operation should we write the journal sb 1663 * 1664 * Update a journal's superblock information about log tail and write it to 1665 * disk, waiting for the IO to complete. 1666 */ 1667 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, 1668 unsigned long tail_block, int write_op) 1669 { 1670 journal_superblock_t *sb = journal->j_superblock; 1671 int ret; 1672 1673 if (is_journal_aborted(journal)) 1674 return -EIO; 1675 if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) { 1676 jbd2_journal_abort(journal, -EIO); 1677 return -EIO; 1678 } 1679 1680 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1681 jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", 1682 tail_block, tail_tid); 1683 1684 lock_buffer(journal->j_sb_buffer); 1685 sb->s_sequence = cpu_to_be32(tail_tid); 1686 sb->s_start = cpu_to_be32(tail_block); 1687 1688 ret = jbd2_write_superblock(journal, write_op); 1689 if (ret) 1690 goto out; 1691 1692 /* Log is no longer empty */ 1693 write_lock(&journal->j_state_lock); 1694 WARN_ON(!sb->s_sequence); 1695 journal->j_flags &= ~JBD2_FLUSHED; 1696 write_unlock(&journal->j_state_lock); 1697 1698 out: 1699 return ret; 1700 } 1701 1702 /** 1703 * jbd2_mark_journal_empty() - Mark on disk journal as empty. 1704 * @journal: The journal to update. 1705 * @write_op: With which operation should we write the journal sb 1706 * 1707 * Update a journal's dynamic superblock fields to show that journal is empty. 1708 * Write updated superblock to disk waiting for IO to complete. 1709 */ 1710 static void jbd2_mark_journal_empty(journal_t *journal, int write_op) 1711 { 1712 journal_superblock_t *sb = journal->j_superblock; 1713 bool had_fast_commit = false; 1714 1715 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1716 lock_buffer(journal->j_sb_buffer); 1717 if (sb->s_start == 0) { /* Is it already empty? */ 1718 unlock_buffer(journal->j_sb_buffer); 1719 return; 1720 } 1721 1722 jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n", 1723 journal->j_tail_sequence); 1724 1725 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); 1726 sb->s_start = cpu_to_be32(0); 1727 if (jbd2_has_feature_fast_commit(journal)) { 1728 /* 1729 * When journal is clean, no need to commit fast commit flag and 1730 * make file system incompatible with older kernels. 1731 */ 1732 jbd2_clear_feature_fast_commit(journal); 1733 had_fast_commit = true; 1734 } 1735 1736 jbd2_write_superblock(journal, write_op); 1737 1738 if (had_fast_commit) 1739 jbd2_set_feature_fast_commit(journal); 1740 1741 /* Log is no longer empty */ 1742 write_lock(&journal->j_state_lock); 1743 journal->j_flags |= JBD2_FLUSHED; 1744 write_unlock(&journal->j_state_lock); 1745 } 1746 1747 /** 1748 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock) 1749 * @journal: The journal to erase. 1750 * @flags: A discard/zeroout request is sent for each physically contigous 1751 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or 1752 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation 1753 * to perform. 1754 * 1755 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes 1756 * will be explicitly written if no hardware offload is available, see 1757 * blkdev_issue_zeroout for more details. 1758 */ 1759 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags) 1760 { 1761 int err = 0; 1762 unsigned long block, log_offset; /* logical */ 1763 unsigned long long phys_block, block_start, block_stop; /* physical */ 1764 loff_t byte_start, byte_stop, byte_count; 1765 struct request_queue *q = bdev_get_queue(journal->j_dev); 1766 1767 /* flags must be set to either discard or zeroout */ 1768 if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags || 1769 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && 1770 (flags & JBD2_JOURNAL_FLUSH_ZEROOUT))) 1771 return -EINVAL; 1772 1773 if (!q) 1774 return -ENXIO; 1775 1776 if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && !blk_queue_discard(q)) 1777 return -EOPNOTSUPP; 1778 1779 /* 1780 * lookup block mapping and issue discard/zeroout for each 1781 * contiguous region 1782 */ 1783 log_offset = be32_to_cpu(journal->j_superblock->s_first); 1784 block_start = ~0ULL; 1785 for (block = log_offset; block < journal->j_total_len; block++) { 1786 err = jbd2_journal_bmap(journal, block, &phys_block); 1787 if (err) { 1788 pr_err("JBD2: bad block at offset %lu", block); 1789 return err; 1790 } 1791 1792 if (block_start == ~0ULL) { 1793 block_start = phys_block; 1794 block_stop = block_start - 1; 1795 } 1796 1797 /* 1798 * last block not contiguous with current block, 1799 * process last contiguous region and return to this block on 1800 * next loop 1801 */ 1802 if (phys_block != block_stop + 1) { 1803 block--; 1804 } else { 1805 block_stop++; 1806 /* 1807 * if this isn't the last block of journal, 1808 * no need to process now because next block may also 1809 * be part of this contiguous region 1810 */ 1811 if (block != journal->j_total_len - 1) 1812 continue; 1813 } 1814 1815 /* 1816 * end of contiguous region or this is last block of journal, 1817 * take care of the region 1818 */ 1819 byte_start = block_start * journal->j_blocksize; 1820 byte_stop = block_stop * journal->j_blocksize; 1821 byte_count = (block_stop - block_start + 1) * 1822 journal->j_blocksize; 1823 1824 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping, 1825 byte_start, byte_stop); 1826 1827 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) { 1828 err = blkdev_issue_discard(journal->j_dev, 1829 byte_start >> SECTOR_SHIFT, 1830 byte_count >> SECTOR_SHIFT, 1831 GFP_NOFS, 0); 1832 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) { 1833 err = blkdev_issue_zeroout(journal->j_dev, 1834 byte_start >> SECTOR_SHIFT, 1835 byte_count >> SECTOR_SHIFT, 1836 GFP_NOFS, 0); 1837 } 1838 1839 if (unlikely(err != 0)) { 1840 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu", 1841 err, block_start, block_stop); 1842 return err; 1843 } 1844 1845 /* reset start and stop after processing a region */ 1846 block_start = ~0ULL; 1847 } 1848 1849 return blkdev_issue_flush(journal->j_dev); 1850 } 1851 1852 /** 1853 * jbd2_journal_update_sb_errno() - Update error in the journal. 1854 * @journal: The journal to update. 1855 * 1856 * Update a journal's errno. Write updated superblock to disk waiting for IO 1857 * to complete. 1858 */ 1859 void jbd2_journal_update_sb_errno(journal_t *journal) 1860 { 1861 journal_superblock_t *sb = journal->j_superblock; 1862 int errcode; 1863 1864 lock_buffer(journal->j_sb_buffer); 1865 errcode = journal->j_errno; 1866 if (errcode == -ESHUTDOWN) 1867 errcode = 0; 1868 jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode); 1869 sb->s_errno = cpu_to_be32(errcode); 1870 1871 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA); 1872 } 1873 EXPORT_SYMBOL(jbd2_journal_update_sb_errno); 1874 1875 static int journal_revoke_records_per_block(journal_t *journal) 1876 { 1877 int record_size; 1878 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t); 1879 1880 if (jbd2_has_feature_64bit(journal)) 1881 record_size = 8; 1882 else 1883 record_size = 4; 1884 1885 if (jbd2_journal_has_csum_v2or3(journal)) 1886 space -= sizeof(struct jbd2_journal_block_tail); 1887 return space / record_size; 1888 } 1889 1890 /* 1891 * Read the superblock for a given journal, performing initial 1892 * validation of the format. 1893 */ 1894 static int journal_get_superblock(journal_t *journal) 1895 { 1896 struct buffer_head *bh; 1897 journal_superblock_t *sb; 1898 int err = -EIO; 1899 1900 bh = journal->j_sb_buffer; 1901 1902 J_ASSERT(bh != NULL); 1903 if (!buffer_uptodate(bh)) { 1904 ll_rw_block(REQ_OP_READ, 0, 1, &bh); 1905 wait_on_buffer(bh); 1906 if (!buffer_uptodate(bh)) { 1907 printk(KERN_ERR 1908 "JBD2: IO error reading journal superblock\n"); 1909 goto out; 1910 } 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 jbd_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 jbd_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 jbd_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_page && bh->b_page->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