1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * linux/fs/jbd2/journal.c 4 * 5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 6 * 7 * Copyright 1998 Red Hat corp --- All Rights Reserved 8 * 9 * Generic filesystem journal-writing code; part of the ext2fs 10 * journaling system. 11 * 12 * This file manages journals: areas of disk reserved for logging 13 * transactional updates. This includes the kernel journaling thread 14 * which is responsible for scheduling updates to the log. 15 * 16 * We do not actually manage the physical storage of the journal in this 17 * file: that is left to a per-journal policy function, which allows us 18 * to store the journal within a filesystem-specified area for ext2 19 * journaling (ext2 can use a reserved inode for storing the log). 20 */ 21 22 #include <linux/module.h> 23 #include <linux/time.h> 24 #include <linux/fs.h> 25 #include <linux/jbd2.h> 26 #include <linux/errno.h> 27 #include <linux/slab.h> 28 #include <linux/init.h> 29 #include <linux/mm.h> 30 #include <linux/freezer.h> 31 #include <linux/pagemap.h> 32 #include <linux/kthread.h> 33 #include <linux/poison.h> 34 #include <linux/proc_fs.h> 35 #include <linux/seq_file.h> 36 #include <linux/math64.h> 37 #include <linux/hash.h> 38 #include <linux/log2.h> 39 #include <linux/vmalloc.h> 40 #include <linux/backing-dev.h> 41 #include <linux/bitops.h> 42 #include <linux/ratelimit.h> 43 #include <linux/sched/mm.h> 44 45 #define CREATE_TRACE_POINTS 46 #include <trace/events/jbd2.h> 47 48 #include <linux/uaccess.h> 49 #include <asm/page.h> 50 51 #ifdef CONFIG_JBD2_DEBUG 52 static ushort jbd2_journal_enable_debug __read_mostly; 53 54 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644); 55 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2"); 56 #endif 57 58 EXPORT_SYMBOL(jbd2_journal_extend); 59 EXPORT_SYMBOL(jbd2_journal_stop); 60 EXPORT_SYMBOL(jbd2_journal_lock_updates); 61 EXPORT_SYMBOL(jbd2_journal_unlock_updates); 62 EXPORT_SYMBOL(jbd2_journal_get_write_access); 63 EXPORT_SYMBOL(jbd2_journal_get_create_access); 64 EXPORT_SYMBOL(jbd2_journal_get_undo_access); 65 EXPORT_SYMBOL(jbd2_journal_set_triggers); 66 EXPORT_SYMBOL(jbd2_journal_dirty_metadata); 67 EXPORT_SYMBOL(jbd2_journal_forget); 68 EXPORT_SYMBOL(jbd2_journal_flush); 69 EXPORT_SYMBOL(jbd2_journal_revoke); 70 71 EXPORT_SYMBOL(jbd2_journal_init_dev); 72 EXPORT_SYMBOL(jbd2_journal_init_inode); 73 EXPORT_SYMBOL(jbd2_journal_check_used_features); 74 EXPORT_SYMBOL(jbd2_journal_check_available_features); 75 EXPORT_SYMBOL(jbd2_journal_set_features); 76 EXPORT_SYMBOL(jbd2_journal_load); 77 EXPORT_SYMBOL(jbd2_journal_destroy); 78 EXPORT_SYMBOL(jbd2_journal_abort); 79 EXPORT_SYMBOL(jbd2_journal_errno); 80 EXPORT_SYMBOL(jbd2_journal_ack_err); 81 EXPORT_SYMBOL(jbd2_journal_clear_err); 82 EXPORT_SYMBOL(jbd2_log_wait_commit); 83 EXPORT_SYMBOL(jbd2_journal_start_commit); 84 EXPORT_SYMBOL(jbd2_journal_force_commit_nested); 85 EXPORT_SYMBOL(jbd2_journal_wipe); 86 EXPORT_SYMBOL(jbd2_journal_blocks_per_page); 87 EXPORT_SYMBOL(jbd2_journal_invalidate_folio); 88 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); 89 EXPORT_SYMBOL(jbd2_journal_force_commit); 90 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write); 91 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait); 92 EXPORT_SYMBOL(jbd2_journal_submit_inode_data_buffers); 93 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers); 94 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); 95 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); 96 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); 97 EXPORT_SYMBOL(jbd2_inode_cache); 98 99 static int jbd2_journal_create_slab(size_t slab_size); 100 101 #ifdef CONFIG_JBD2_DEBUG 102 void __jbd2_debug(int level, const char *file, const char *func, 103 unsigned int line, const char *fmt, ...) 104 { 105 struct va_format vaf; 106 va_list args; 107 108 if (level > jbd2_journal_enable_debug) 109 return; 110 va_start(args, fmt); 111 vaf.fmt = fmt; 112 vaf.va = &args; 113 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf); 114 va_end(args); 115 } 116 #endif 117 118 /* Checksumming functions */ 119 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) 120 { 121 if (!jbd2_journal_has_csum_v2or3_feature(j)) 122 return 1; 123 124 return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; 125 } 126 127 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) 128 { 129 __u32 csum; 130 __be32 old_csum; 131 132 old_csum = sb->s_checksum; 133 sb->s_checksum = 0; 134 csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); 135 sb->s_checksum = old_csum; 136 137 return cpu_to_be32(csum); 138 } 139 140 /* 141 * Helper function used to manage commit timeouts 142 */ 143 144 static void commit_timeout(struct timer_list *t) 145 { 146 journal_t *journal = from_timer(journal, t, j_commit_timer); 147 148 wake_up_process(journal->j_task); 149 } 150 151 /* 152 * kjournald2: The main thread function used to manage a logging device 153 * journal. 154 * 155 * This kernel thread is responsible for two things: 156 * 157 * 1) COMMIT: Every so often we need to commit the current state of the 158 * filesystem to disk. The journal thread is responsible for writing 159 * all of the metadata buffers to disk. If a fast commit is ongoing 160 * journal thread waits until it's done and then continues from 161 * there on. 162 * 163 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all 164 * of the data in that part of the log has been rewritten elsewhere on 165 * the disk. Flushing these old buffers to reclaim space in the log is 166 * known as checkpointing, and this thread is responsible for that job. 167 */ 168 169 static int kjournald2(void *arg) 170 { 171 journal_t *journal = arg; 172 transaction_t *transaction; 173 174 /* 175 * Set up an interval timer which can be used to trigger a commit wakeup 176 * after the commit interval expires 177 */ 178 timer_setup(&journal->j_commit_timer, commit_timeout, 0); 179 180 set_freezable(); 181 182 /* Record that the journal thread is running */ 183 journal->j_task = current; 184 wake_up(&journal->j_wait_done_commit); 185 186 /* 187 * Make sure that no allocations from this kernel thread will ever 188 * recurse to the fs layer because we are responsible for the 189 * transaction commit and any fs involvement might get stuck waiting for 190 * the trasn. commit. 191 */ 192 memalloc_nofs_save(); 193 194 /* 195 * And now, wait forever for commit wakeup events. 196 */ 197 write_lock(&journal->j_state_lock); 198 199 loop: 200 if (journal->j_flags & JBD2_UNMOUNT) 201 goto end_loop; 202 203 jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n", 204 journal->j_commit_sequence, journal->j_commit_request); 205 206 if (journal->j_commit_sequence != journal->j_commit_request) { 207 jbd2_debug(1, "OK, requests differ\n"); 208 write_unlock(&journal->j_state_lock); 209 del_timer_sync(&journal->j_commit_timer); 210 jbd2_journal_commit_transaction(journal); 211 write_lock(&journal->j_state_lock); 212 goto loop; 213 } 214 215 wake_up(&journal->j_wait_done_commit); 216 if (freezing(current)) { 217 /* 218 * The simpler the better. Flushing journal isn't a 219 * good idea, because that depends on threads that may 220 * be already stopped. 221 */ 222 jbd2_debug(1, "Now suspending kjournald2\n"); 223 write_unlock(&journal->j_state_lock); 224 try_to_freeze(); 225 write_lock(&journal->j_state_lock); 226 } else { 227 /* 228 * We assume on resume that commits are already there, 229 * so we don't sleep 230 */ 231 DEFINE_WAIT(wait); 232 int should_sleep = 1; 233 234 prepare_to_wait(&journal->j_wait_commit, &wait, 235 TASK_INTERRUPTIBLE); 236 if (journal->j_commit_sequence != journal->j_commit_request) 237 should_sleep = 0; 238 transaction = journal->j_running_transaction; 239 if (transaction && time_after_eq(jiffies, 240 transaction->t_expires)) 241 should_sleep = 0; 242 if (journal->j_flags & JBD2_UNMOUNT) 243 should_sleep = 0; 244 if (should_sleep) { 245 write_unlock(&journal->j_state_lock); 246 schedule(); 247 write_lock(&journal->j_state_lock); 248 } 249 finish_wait(&journal->j_wait_commit, &wait); 250 } 251 252 jbd2_debug(1, "kjournald2 wakes\n"); 253 254 /* 255 * Were we woken up by a commit wakeup event? 256 */ 257 transaction = journal->j_running_transaction; 258 if (transaction && time_after_eq(jiffies, transaction->t_expires)) { 259 journal->j_commit_request = transaction->t_tid; 260 jbd2_debug(1, "woke because of timeout\n"); 261 } 262 goto loop; 263 264 end_loop: 265 del_timer_sync(&journal->j_commit_timer); 266 journal->j_task = NULL; 267 wake_up(&journal->j_wait_done_commit); 268 jbd2_debug(1, "Journal thread exiting.\n"); 269 write_unlock(&journal->j_state_lock); 270 return 0; 271 } 272 273 static int jbd2_journal_start_thread(journal_t *journal) 274 { 275 struct task_struct *t; 276 277 t = kthread_run(kjournald2, journal, "jbd2/%s", 278 journal->j_devname); 279 if (IS_ERR(t)) 280 return PTR_ERR(t); 281 282 wait_event(journal->j_wait_done_commit, journal->j_task != NULL); 283 return 0; 284 } 285 286 static void journal_kill_thread(journal_t *journal) 287 { 288 write_lock(&journal->j_state_lock); 289 journal->j_flags |= JBD2_UNMOUNT; 290 291 while (journal->j_task) { 292 write_unlock(&journal->j_state_lock); 293 wake_up(&journal->j_wait_commit); 294 wait_event(journal->j_wait_done_commit, journal->j_task == NULL); 295 write_lock(&journal->j_state_lock); 296 } 297 write_unlock(&journal->j_state_lock); 298 } 299 300 /* 301 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. 302 * 303 * Writes a metadata buffer to a given disk block. The actual IO is not 304 * performed but a new buffer_head is constructed which labels the data 305 * to be written with the correct destination disk block. 306 * 307 * Any magic-number escaping which needs to be done will cause a 308 * copy-out here. If the buffer happens to start with the 309 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the 310 * magic number is only written to the log for descripter blocks. In 311 * this case, we copy the data and replace the first word with 0, and we 312 * return a result code which indicates that this buffer needs to be 313 * marked as an escaped buffer in the corresponding log descriptor 314 * block. The missing word can then be restored when the block is read 315 * during recovery. 316 * 317 * If the source buffer has already been modified by a new transaction 318 * since we took the last commit snapshot, we use the frozen copy of 319 * that data for IO. If we end up using the existing buffer_head's data 320 * for the write, then we have to make sure nobody modifies it while the 321 * IO is in progress. do_get_write_access() handles this. 322 * 323 * The function returns a pointer to the buffer_head to be used for IO. 324 * 325 * 326 * Return value: 327 * <0: Error 328 * >=0: Finished OK 329 * 330 * On success: 331 * Bit 0 set == escape performed on the data 332 * Bit 1 set == buffer copy-out performed (kfree the data after IO) 333 */ 334 335 int jbd2_journal_write_metadata_buffer(transaction_t *transaction, 336 struct journal_head *jh_in, 337 struct buffer_head **bh_out, 338 sector_t blocknr) 339 { 340 int need_copy_out = 0; 341 int done_copy_out = 0; 342 int do_escape = 0; 343 char *mapped_data; 344 struct buffer_head *new_bh; 345 struct page *new_page; 346 unsigned int new_offset; 347 struct buffer_head *bh_in = jh2bh(jh_in); 348 journal_t *journal = transaction->t_journal; 349 350 /* 351 * The buffer really shouldn't be locked: only the current committing 352 * transaction is allowed to write it, so nobody else is allowed 353 * to do any IO. 354 * 355 * akpm: except if we're journalling data, and write() output is 356 * also part of a shared mapping, and another thread has 357 * decided to launch a writepage() against this buffer. 358 */ 359 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); 360 361 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); 362 363 /* keep subsequent assertions sane */ 364 atomic_set(&new_bh->b_count, 1); 365 366 spin_lock(&jh_in->b_state_lock); 367 repeat: 368 /* 369 * If a new transaction has already done a buffer copy-out, then 370 * we use that version of the data for the commit. 371 */ 372 if (jh_in->b_frozen_data) { 373 done_copy_out = 1; 374 new_page = virt_to_page(jh_in->b_frozen_data); 375 new_offset = offset_in_page(jh_in->b_frozen_data); 376 } else { 377 new_page = jh2bh(jh_in)->b_page; 378 new_offset = offset_in_page(jh2bh(jh_in)->b_data); 379 } 380 381 mapped_data = kmap_atomic(new_page); 382 /* 383 * Fire data frozen trigger if data already wasn't frozen. Do this 384 * before checking for escaping, as the trigger may modify the magic 385 * offset. If a copy-out happens afterwards, it will have the correct 386 * data in the buffer. 387 */ 388 if (!done_copy_out) 389 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, 390 jh_in->b_triggers); 391 392 /* 393 * Check for escaping 394 */ 395 if (*((__be32 *)(mapped_data + new_offset)) == 396 cpu_to_be32(JBD2_MAGIC_NUMBER)) { 397 need_copy_out = 1; 398 do_escape = 1; 399 } 400 kunmap_atomic(mapped_data); 401 402 /* 403 * Do we need to do a data copy? 404 */ 405 if (need_copy_out && !done_copy_out) { 406 char *tmp; 407 408 spin_unlock(&jh_in->b_state_lock); 409 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); 410 if (!tmp) { 411 brelse(new_bh); 412 return -ENOMEM; 413 } 414 spin_lock(&jh_in->b_state_lock); 415 if (jh_in->b_frozen_data) { 416 jbd2_free(tmp, bh_in->b_size); 417 goto repeat; 418 } 419 420 jh_in->b_frozen_data = tmp; 421 mapped_data = kmap_atomic(new_page); 422 memcpy(tmp, mapped_data + new_offset, bh_in->b_size); 423 kunmap_atomic(mapped_data); 424 425 new_page = virt_to_page(tmp); 426 new_offset = offset_in_page(tmp); 427 done_copy_out = 1; 428 429 /* 430 * This isn't strictly necessary, as we're using frozen 431 * data for the escaping, but it keeps consistency with 432 * b_frozen_data usage. 433 */ 434 jh_in->b_frozen_triggers = jh_in->b_triggers; 435 } 436 437 /* 438 * Did we need to do an escaping? Now we've done all the 439 * copying, we can finally do so. 440 */ 441 if (do_escape) { 442 mapped_data = kmap_atomic(new_page); 443 *((unsigned int *)(mapped_data + new_offset)) = 0; 444 kunmap_atomic(mapped_data); 445 } 446 447 set_bh_page(new_bh, new_page, new_offset); 448 new_bh->b_size = bh_in->b_size; 449 new_bh->b_bdev = journal->j_dev; 450 new_bh->b_blocknr = blocknr; 451 new_bh->b_private = bh_in; 452 set_buffer_mapped(new_bh); 453 set_buffer_dirty(new_bh); 454 455 *bh_out = new_bh; 456 457 /* 458 * The to-be-written buffer needs to get moved to the io queue, 459 * and the original buffer whose contents we are shadowing or 460 * copying is moved to the transaction's shadow queue. 461 */ 462 JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); 463 spin_lock(&journal->j_list_lock); 464 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); 465 spin_unlock(&journal->j_list_lock); 466 set_buffer_shadow(bh_in); 467 spin_unlock(&jh_in->b_state_lock); 468 469 return do_escape | (done_copy_out << 1); 470 } 471 472 /* 473 * Allocation code for the journal file. Manage the space left in the 474 * journal, so that we can begin checkpointing when appropriate. 475 */ 476 477 /* 478 * Called with j_state_lock locked for writing. 479 * Returns true if a transaction commit was started. 480 */ 481 static int __jbd2_log_start_commit(journal_t *journal, tid_t target) 482 { 483 /* Return if the txn has already requested to be committed */ 484 if (journal->j_commit_request == target) 485 return 0; 486 487 /* 488 * The only transaction we can possibly wait upon is the 489 * currently running transaction (if it exists). Otherwise, 490 * the target tid must be an old one. 491 */ 492 if (journal->j_running_transaction && 493 journal->j_running_transaction->t_tid == target) { 494 /* 495 * We want a new commit: OK, mark the request and wakeup the 496 * commit thread. We do _not_ do the commit ourselves. 497 */ 498 499 journal->j_commit_request = target; 500 jbd2_debug(1, "JBD2: requesting commit %u/%u\n", 501 journal->j_commit_request, 502 journal->j_commit_sequence); 503 journal->j_running_transaction->t_requested = jiffies; 504 wake_up(&journal->j_wait_commit); 505 return 1; 506 } else if (!tid_geq(journal->j_commit_request, target)) 507 /* This should never happen, but if it does, preserve 508 the evidence before kjournald goes into a loop and 509 increments j_commit_sequence beyond all recognition. */ 510 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", 511 journal->j_commit_request, 512 journal->j_commit_sequence, 513 target, journal->j_running_transaction ? 514 journal->j_running_transaction->t_tid : 0); 515 return 0; 516 } 517 518 int jbd2_log_start_commit(journal_t *journal, tid_t tid) 519 { 520 int ret; 521 522 write_lock(&journal->j_state_lock); 523 ret = __jbd2_log_start_commit(journal, tid); 524 write_unlock(&journal->j_state_lock); 525 return ret; 526 } 527 528 /* 529 * Force and wait any uncommitted transactions. We can only force the running 530 * transaction if we don't have an active handle, otherwise, we will deadlock. 531 * Returns: <0 in case of error, 532 * 0 if nothing to commit, 533 * 1 if transaction was successfully committed. 534 */ 535 static int __jbd2_journal_force_commit(journal_t *journal) 536 { 537 transaction_t *transaction = NULL; 538 tid_t tid; 539 int need_to_start = 0, ret = 0; 540 541 read_lock(&journal->j_state_lock); 542 if (journal->j_running_transaction && !current->journal_info) { 543 transaction = journal->j_running_transaction; 544 if (!tid_geq(journal->j_commit_request, transaction->t_tid)) 545 need_to_start = 1; 546 } else if (journal->j_committing_transaction) 547 transaction = journal->j_committing_transaction; 548 549 if (!transaction) { 550 /* Nothing to commit */ 551 read_unlock(&journal->j_state_lock); 552 return 0; 553 } 554 tid = transaction->t_tid; 555 read_unlock(&journal->j_state_lock); 556 if (need_to_start) 557 jbd2_log_start_commit(journal, tid); 558 ret = jbd2_log_wait_commit(journal, tid); 559 if (!ret) 560 ret = 1; 561 562 return ret; 563 } 564 565 /** 566 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the 567 * calling process is not within transaction. 568 * 569 * @journal: journal to force 570 * Returns true if progress was made. 571 * 572 * This is used for forcing out undo-protected data which contains 573 * bitmaps, when the fs is running out of space. 574 */ 575 int jbd2_journal_force_commit_nested(journal_t *journal) 576 { 577 int ret; 578 579 ret = __jbd2_journal_force_commit(journal); 580 return ret > 0; 581 } 582 583 /** 584 * jbd2_journal_force_commit() - force any uncommitted transactions 585 * @journal: journal to force 586 * 587 * Caller want unconditional commit. We can only force the running transaction 588 * if we don't have an active handle, otherwise, we will deadlock. 589 */ 590 int jbd2_journal_force_commit(journal_t *journal) 591 { 592 int ret; 593 594 J_ASSERT(!current->journal_info); 595 ret = __jbd2_journal_force_commit(journal); 596 if (ret > 0) 597 ret = 0; 598 return ret; 599 } 600 601 /* 602 * Start a commit of the current running transaction (if any). Returns true 603 * if a transaction is going to be committed (or is currently already 604 * committing), and fills its tid in at *ptid 605 */ 606 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) 607 { 608 int ret = 0; 609 610 write_lock(&journal->j_state_lock); 611 if (journal->j_running_transaction) { 612 tid_t tid = journal->j_running_transaction->t_tid; 613 614 __jbd2_log_start_commit(journal, tid); 615 /* There's a running transaction and we've just made sure 616 * it's commit has been scheduled. */ 617 if (ptid) 618 *ptid = tid; 619 ret = 1; 620 } else if (journal->j_committing_transaction) { 621 /* 622 * If commit has been started, then we have to wait for 623 * completion of that transaction. 624 */ 625 if (ptid) 626 *ptid = journal->j_committing_transaction->t_tid; 627 ret = 1; 628 } 629 write_unlock(&journal->j_state_lock); 630 return ret; 631 } 632 633 /* 634 * Return 1 if a given transaction has not yet sent barrier request 635 * connected with a transaction commit. If 0 is returned, transaction 636 * may or may not have sent the barrier. Used to avoid sending barrier 637 * twice in common cases. 638 */ 639 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) 640 { 641 int ret = 0; 642 transaction_t *commit_trans; 643 644 if (!(journal->j_flags & JBD2_BARRIER)) 645 return 0; 646 read_lock(&journal->j_state_lock); 647 /* Transaction already committed? */ 648 if (tid_geq(journal->j_commit_sequence, tid)) 649 goto out; 650 commit_trans = journal->j_committing_transaction; 651 if (!commit_trans || commit_trans->t_tid != tid) { 652 ret = 1; 653 goto out; 654 } 655 /* 656 * Transaction is being committed and we already proceeded to 657 * submitting a flush to fs partition? 658 */ 659 if (journal->j_fs_dev != journal->j_dev) { 660 if (!commit_trans->t_need_data_flush || 661 commit_trans->t_state >= T_COMMIT_DFLUSH) 662 goto out; 663 } else { 664 if (commit_trans->t_state >= T_COMMIT_JFLUSH) 665 goto out; 666 } 667 ret = 1; 668 out: 669 read_unlock(&journal->j_state_lock); 670 return ret; 671 } 672 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); 673 674 /* 675 * Wait for a specified commit to complete. 676 * The caller may not hold the journal lock. 677 */ 678 int jbd2_log_wait_commit(journal_t *journal, tid_t tid) 679 { 680 int err = 0; 681 682 read_lock(&journal->j_state_lock); 683 #ifdef CONFIG_PROVE_LOCKING 684 /* 685 * Some callers make sure transaction is already committing and in that 686 * case we cannot block on open handles anymore. So don't warn in that 687 * case. 688 */ 689 if (tid_gt(tid, journal->j_commit_sequence) && 690 (!journal->j_committing_transaction || 691 journal->j_committing_transaction->t_tid != tid)) { 692 read_unlock(&journal->j_state_lock); 693 jbd2_might_wait_for_commit(journal); 694 read_lock(&journal->j_state_lock); 695 } 696 #endif 697 #ifdef CONFIG_JBD2_DEBUG 698 if (!tid_geq(journal->j_commit_request, tid)) { 699 printk(KERN_ERR 700 "%s: error: j_commit_request=%u, tid=%u\n", 701 __func__, journal->j_commit_request, tid); 702 } 703 #endif 704 while (tid_gt(tid, journal->j_commit_sequence)) { 705 jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n", 706 tid, journal->j_commit_sequence); 707 read_unlock(&journal->j_state_lock); 708 wake_up(&journal->j_wait_commit); 709 wait_event(journal->j_wait_done_commit, 710 !tid_gt(tid, journal->j_commit_sequence)); 711 read_lock(&journal->j_state_lock); 712 } 713 read_unlock(&journal->j_state_lock); 714 715 if (unlikely(is_journal_aborted(journal))) 716 err = -EIO; 717 return err; 718 } 719 720 /* 721 * Start a fast commit. If there's an ongoing fast or full commit wait for 722 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY 723 * if a fast commit is not needed, either because there's an already a commit 724 * going on or this tid has already been committed. Returns -EINVAL if no jbd2 725 * commit has yet been performed. 726 */ 727 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid) 728 { 729 if (unlikely(is_journal_aborted(journal))) 730 return -EIO; 731 /* 732 * Fast commits only allowed if at least one full commit has 733 * been processed. 734 */ 735 if (!journal->j_stats.ts_tid) 736 return -EINVAL; 737 738 write_lock(&journal->j_state_lock); 739 if (tid <= journal->j_commit_sequence) { 740 write_unlock(&journal->j_state_lock); 741 return -EALREADY; 742 } 743 744 if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING || 745 (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) { 746 DEFINE_WAIT(wait); 747 748 prepare_to_wait(&journal->j_fc_wait, &wait, 749 TASK_UNINTERRUPTIBLE); 750 write_unlock(&journal->j_state_lock); 751 schedule(); 752 finish_wait(&journal->j_fc_wait, &wait); 753 return -EALREADY; 754 } 755 journal->j_flags |= JBD2_FAST_COMMIT_ONGOING; 756 write_unlock(&journal->j_state_lock); 757 jbd2_journal_lock_updates(journal); 758 759 return 0; 760 } 761 EXPORT_SYMBOL(jbd2_fc_begin_commit); 762 763 /* 764 * Stop a fast commit. If fallback is set, this function starts commit of 765 * TID tid before any other fast commit can start. 766 */ 767 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback) 768 { 769 jbd2_journal_unlock_updates(journal); 770 if (journal->j_fc_cleanup_callback) 771 journal->j_fc_cleanup_callback(journal, 0, tid); 772 write_lock(&journal->j_state_lock); 773 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING; 774 if (fallback) 775 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING; 776 write_unlock(&journal->j_state_lock); 777 wake_up(&journal->j_fc_wait); 778 if (fallback) 779 return jbd2_complete_transaction(journal, tid); 780 return 0; 781 } 782 783 int jbd2_fc_end_commit(journal_t *journal) 784 { 785 return __jbd2_fc_end_commit(journal, 0, false); 786 } 787 EXPORT_SYMBOL(jbd2_fc_end_commit); 788 789 int jbd2_fc_end_commit_fallback(journal_t *journal) 790 { 791 tid_t tid; 792 793 read_lock(&journal->j_state_lock); 794 tid = journal->j_running_transaction ? 795 journal->j_running_transaction->t_tid : 0; 796 read_unlock(&journal->j_state_lock); 797 return __jbd2_fc_end_commit(journal, tid, true); 798 } 799 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback); 800 801 /* Return 1 when transaction with given tid has already committed. */ 802 int jbd2_transaction_committed(journal_t *journal, tid_t tid) 803 { 804 int ret = 1; 805 806 read_lock(&journal->j_state_lock); 807 if (journal->j_running_transaction && 808 journal->j_running_transaction->t_tid == tid) 809 ret = 0; 810 if (journal->j_committing_transaction && 811 journal->j_committing_transaction->t_tid == tid) 812 ret = 0; 813 read_unlock(&journal->j_state_lock); 814 return ret; 815 } 816 EXPORT_SYMBOL(jbd2_transaction_committed); 817 818 /* 819 * When this function returns the transaction corresponding to tid 820 * will be completed. If the transaction has currently running, start 821 * committing that transaction before waiting for it to complete. If 822 * the transaction id is stale, it is by definition already completed, 823 * so just return SUCCESS. 824 */ 825 int jbd2_complete_transaction(journal_t *journal, tid_t tid) 826 { 827 int need_to_wait = 1; 828 829 read_lock(&journal->j_state_lock); 830 if (journal->j_running_transaction && 831 journal->j_running_transaction->t_tid == tid) { 832 if (journal->j_commit_request != tid) { 833 /* transaction not yet started, so request it */ 834 read_unlock(&journal->j_state_lock); 835 jbd2_log_start_commit(journal, tid); 836 goto wait_commit; 837 } 838 } else if (!(journal->j_committing_transaction && 839 journal->j_committing_transaction->t_tid == tid)) 840 need_to_wait = 0; 841 read_unlock(&journal->j_state_lock); 842 if (!need_to_wait) 843 return 0; 844 wait_commit: 845 return jbd2_log_wait_commit(journal, tid); 846 } 847 EXPORT_SYMBOL(jbd2_complete_transaction); 848 849 /* 850 * Log buffer allocation routines: 851 */ 852 853 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) 854 { 855 unsigned long blocknr; 856 857 write_lock(&journal->j_state_lock); 858 J_ASSERT(journal->j_free > 1); 859 860 blocknr = journal->j_head; 861 journal->j_head++; 862 journal->j_free--; 863 if (journal->j_head == journal->j_last) 864 journal->j_head = journal->j_first; 865 write_unlock(&journal->j_state_lock); 866 return jbd2_journal_bmap(journal, blocknr, retp); 867 } 868 869 /* Map one fast commit buffer for use by the file system */ 870 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out) 871 { 872 unsigned long long pblock; 873 unsigned long blocknr; 874 int ret = 0; 875 struct buffer_head *bh; 876 int fc_off; 877 878 *bh_out = NULL; 879 880 if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) { 881 fc_off = journal->j_fc_off; 882 blocknr = journal->j_fc_first + fc_off; 883 journal->j_fc_off++; 884 } else { 885 ret = -EINVAL; 886 } 887 888 if (ret) 889 return ret; 890 891 ret = jbd2_journal_bmap(journal, blocknr, &pblock); 892 if (ret) 893 return ret; 894 895 bh = __getblk(journal->j_dev, pblock, journal->j_blocksize); 896 if (!bh) 897 return -ENOMEM; 898 899 900 journal->j_fc_wbuf[fc_off] = bh; 901 902 *bh_out = bh; 903 904 return 0; 905 } 906 EXPORT_SYMBOL(jbd2_fc_get_buf); 907 908 /* 909 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf 910 * for completion. 911 */ 912 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks) 913 { 914 struct buffer_head *bh; 915 int i, j_fc_off; 916 917 j_fc_off = journal->j_fc_off; 918 919 /* 920 * Wait in reverse order to minimize chances of us being woken up before 921 * all IOs have completed 922 */ 923 for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) { 924 bh = journal->j_fc_wbuf[i]; 925 wait_on_buffer(bh); 926 put_bh(bh); 927 journal->j_fc_wbuf[i] = NULL; 928 if (unlikely(!buffer_uptodate(bh))) 929 return -EIO; 930 } 931 932 return 0; 933 } 934 EXPORT_SYMBOL(jbd2_fc_wait_bufs); 935 936 int jbd2_fc_release_bufs(journal_t *journal) 937 { 938 struct buffer_head *bh; 939 int i, j_fc_off; 940 941 j_fc_off = journal->j_fc_off; 942 943 for (i = j_fc_off - 1; i >= 0; i--) { 944 bh = journal->j_fc_wbuf[i]; 945 if (!bh) 946 break; 947 put_bh(bh); 948 journal->j_fc_wbuf[i] = NULL; 949 } 950 951 return 0; 952 } 953 EXPORT_SYMBOL(jbd2_fc_release_bufs); 954 955 /* 956 * Conversion of logical to physical block numbers for the journal 957 * 958 * On external journals the journal blocks are identity-mapped, so 959 * this is a no-op. If needed, we can use j_blk_offset - everything is 960 * ready. 961 */ 962 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, 963 unsigned long long *retp) 964 { 965 int err = 0; 966 unsigned long long ret; 967 sector_t block = 0; 968 969 if (journal->j_inode) { 970 block = blocknr; 971 ret = bmap(journal->j_inode, &block); 972 973 if (ret || !block) { 974 printk(KERN_ALERT "%s: journal block not found " 975 "at offset %lu on %s\n", 976 __func__, blocknr, journal->j_devname); 977 err = -EIO; 978 jbd2_journal_abort(journal, err); 979 } else { 980 *retp = block; 981 } 982 983 } else { 984 *retp = blocknr; /* +journal->j_blk_offset */ 985 } 986 return err; 987 } 988 989 /* 990 * We play buffer_head aliasing tricks to write data/metadata blocks to 991 * the journal without copying their contents, but for journal 992 * descriptor blocks we do need to generate bona fide buffers. 993 * 994 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying 995 * the buffer's contents they really should run flush_dcache_page(bh->b_page). 996 * But we don't bother doing that, so there will be coherency problems with 997 * mmaps of blockdevs which hold live JBD-controlled filesystems. 998 */ 999 struct buffer_head * 1000 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type) 1001 { 1002 journal_t *journal = transaction->t_journal; 1003 struct buffer_head *bh; 1004 unsigned long long blocknr; 1005 journal_header_t *header; 1006 int err; 1007 1008 err = jbd2_journal_next_log_block(journal, &blocknr); 1009 1010 if (err) 1011 return NULL; 1012 1013 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); 1014 if (!bh) 1015 return NULL; 1016 atomic_dec(&transaction->t_outstanding_credits); 1017 lock_buffer(bh); 1018 memset(bh->b_data, 0, journal->j_blocksize); 1019 header = (journal_header_t *)bh->b_data; 1020 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); 1021 header->h_blocktype = cpu_to_be32(type); 1022 header->h_sequence = cpu_to_be32(transaction->t_tid); 1023 set_buffer_uptodate(bh); 1024 unlock_buffer(bh); 1025 BUFFER_TRACE(bh, "return this buffer"); 1026 return bh; 1027 } 1028 1029 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh) 1030 { 1031 struct jbd2_journal_block_tail *tail; 1032 __u32 csum; 1033 1034 if (!jbd2_journal_has_csum_v2or3(j)) 1035 return; 1036 1037 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - 1038 sizeof(struct jbd2_journal_block_tail)); 1039 tail->t_checksum = 0; 1040 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); 1041 tail->t_checksum = cpu_to_be32(csum); 1042 } 1043 1044 /* 1045 * Return tid of the oldest transaction in the journal and block in the journal 1046 * where the transaction starts. 1047 * 1048 * If the journal is now empty, return which will be the next transaction ID 1049 * we will write and where will that transaction start. 1050 * 1051 * The return value is 0 if journal tail cannot be pushed any further, 1 if 1052 * it can. 1053 */ 1054 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, 1055 unsigned long *block) 1056 { 1057 transaction_t *transaction; 1058 int ret; 1059 1060 read_lock(&journal->j_state_lock); 1061 spin_lock(&journal->j_list_lock); 1062 transaction = journal->j_checkpoint_transactions; 1063 if (transaction) { 1064 *tid = transaction->t_tid; 1065 *block = transaction->t_log_start; 1066 } else if ((transaction = journal->j_committing_transaction) != NULL) { 1067 *tid = transaction->t_tid; 1068 *block = transaction->t_log_start; 1069 } else if ((transaction = journal->j_running_transaction) != NULL) { 1070 *tid = transaction->t_tid; 1071 *block = journal->j_head; 1072 } else { 1073 *tid = journal->j_transaction_sequence; 1074 *block = journal->j_head; 1075 } 1076 ret = tid_gt(*tid, journal->j_tail_sequence); 1077 spin_unlock(&journal->j_list_lock); 1078 read_unlock(&journal->j_state_lock); 1079 1080 return ret; 1081 } 1082 1083 /* 1084 * Update information in journal structure and in on disk journal superblock 1085 * about log tail. This function does not check whether information passed in 1086 * really pushes log tail further. It's responsibility of the caller to make 1087 * sure provided log tail information is valid (e.g. by holding 1088 * j_checkpoint_mutex all the time between computing log tail and calling this 1089 * function as is the case with jbd2_cleanup_journal_tail()). 1090 * 1091 * Requires j_checkpoint_mutex 1092 */ 1093 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1094 { 1095 unsigned long freed; 1096 int ret; 1097 1098 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1099 1100 /* 1101 * We cannot afford for write to remain in drive's caches since as 1102 * soon as we update j_tail, next transaction can start reusing journal 1103 * space and if we lose sb update during power failure we'd replay 1104 * old transaction with possibly newly overwritten data. 1105 */ 1106 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, 1107 REQ_SYNC | REQ_FUA); 1108 if (ret) 1109 goto out; 1110 1111 write_lock(&journal->j_state_lock); 1112 freed = block - journal->j_tail; 1113 if (block < journal->j_tail) 1114 freed += journal->j_last - journal->j_first; 1115 1116 trace_jbd2_update_log_tail(journal, tid, block, freed); 1117 jbd2_debug(1, 1118 "Cleaning journal tail from %u to %u (offset %lu), " 1119 "freeing %lu\n", 1120 journal->j_tail_sequence, tid, block, freed); 1121 1122 journal->j_free += freed; 1123 journal->j_tail_sequence = tid; 1124 journal->j_tail = block; 1125 write_unlock(&journal->j_state_lock); 1126 1127 out: 1128 return ret; 1129 } 1130 1131 /* 1132 * This is a variation of __jbd2_update_log_tail which checks for validity of 1133 * provided log tail and locks j_checkpoint_mutex. So it is safe against races 1134 * with other threads updating log tail. 1135 */ 1136 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1137 { 1138 mutex_lock_io(&journal->j_checkpoint_mutex); 1139 if (tid_gt(tid, journal->j_tail_sequence)) 1140 __jbd2_update_log_tail(journal, tid, block); 1141 mutex_unlock(&journal->j_checkpoint_mutex); 1142 } 1143 1144 struct jbd2_stats_proc_session { 1145 journal_t *journal; 1146 struct transaction_stats_s *stats; 1147 int start; 1148 int max; 1149 }; 1150 1151 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) 1152 { 1153 return *pos ? NULL : SEQ_START_TOKEN; 1154 } 1155 1156 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) 1157 { 1158 (*pos)++; 1159 return NULL; 1160 } 1161 1162 static int jbd2_seq_info_show(struct seq_file *seq, void *v) 1163 { 1164 struct jbd2_stats_proc_session *s = seq->private; 1165 1166 if (v != SEQ_START_TOKEN) 1167 return 0; 1168 seq_printf(seq, "%lu transactions (%lu requested), " 1169 "each up to %u blocks\n", 1170 s->stats->ts_tid, s->stats->ts_requested, 1171 s->journal->j_max_transaction_buffers); 1172 if (s->stats->ts_tid == 0) 1173 return 0; 1174 seq_printf(seq, "average: \n %ums waiting for transaction\n", 1175 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); 1176 seq_printf(seq, " %ums request delay\n", 1177 (s->stats->ts_requested == 0) ? 0 : 1178 jiffies_to_msecs(s->stats->run.rs_request_delay / 1179 s->stats->ts_requested)); 1180 seq_printf(seq, " %ums running transaction\n", 1181 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); 1182 seq_printf(seq, " %ums transaction was being locked\n", 1183 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); 1184 seq_printf(seq, " %ums flushing data (in ordered mode)\n", 1185 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); 1186 seq_printf(seq, " %ums logging transaction\n", 1187 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); 1188 seq_printf(seq, " %lluus average transaction commit time\n", 1189 div_u64(s->journal->j_average_commit_time, 1000)); 1190 seq_printf(seq, " %lu handles per transaction\n", 1191 s->stats->run.rs_handle_count / s->stats->ts_tid); 1192 seq_printf(seq, " %lu blocks per transaction\n", 1193 s->stats->run.rs_blocks / s->stats->ts_tid); 1194 seq_printf(seq, " %lu logged blocks per transaction\n", 1195 s->stats->run.rs_blocks_logged / s->stats->ts_tid); 1196 return 0; 1197 } 1198 1199 static void jbd2_seq_info_stop(struct seq_file *seq, void *v) 1200 { 1201 } 1202 1203 static const struct seq_operations jbd2_seq_info_ops = { 1204 .start = jbd2_seq_info_start, 1205 .next = jbd2_seq_info_next, 1206 .stop = jbd2_seq_info_stop, 1207 .show = jbd2_seq_info_show, 1208 }; 1209 1210 static int jbd2_seq_info_open(struct inode *inode, struct file *file) 1211 { 1212 journal_t *journal = pde_data(inode); 1213 struct jbd2_stats_proc_session *s; 1214 int rc, size; 1215 1216 s = kmalloc(sizeof(*s), GFP_KERNEL); 1217 if (s == NULL) 1218 return -ENOMEM; 1219 size = sizeof(struct transaction_stats_s); 1220 s->stats = kmalloc(size, GFP_KERNEL); 1221 if (s->stats == NULL) { 1222 kfree(s); 1223 return -ENOMEM; 1224 } 1225 spin_lock(&journal->j_history_lock); 1226 memcpy(s->stats, &journal->j_stats, size); 1227 s->journal = journal; 1228 spin_unlock(&journal->j_history_lock); 1229 1230 rc = seq_open(file, &jbd2_seq_info_ops); 1231 if (rc == 0) { 1232 struct seq_file *m = file->private_data; 1233 m->private = s; 1234 } else { 1235 kfree(s->stats); 1236 kfree(s); 1237 } 1238 return rc; 1239 1240 } 1241 1242 static int jbd2_seq_info_release(struct inode *inode, struct file *file) 1243 { 1244 struct seq_file *seq = file->private_data; 1245 struct jbd2_stats_proc_session *s = seq->private; 1246 kfree(s->stats); 1247 kfree(s); 1248 return seq_release(inode, file); 1249 } 1250 1251 static const struct proc_ops jbd2_info_proc_ops = { 1252 .proc_open = jbd2_seq_info_open, 1253 .proc_read = seq_read, 1254 .proc_lseek = seq_lseek, 1255 .proc_release = jbd2_seq_info_release, 1256 }; 1257 1258 static struct proc_dir_entry *proc_jbd2_stats; 1259 1260 static void jbd2_stats_proc_init(journal_t *journal) 1261 { 1262 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); 1263 if (journal->j_proc_entry) { 1264 proc_create_data("info", S_IRUGO, journal->j_proc_entry, 1265 &jbd2_info_proc_ops, journal); 1266 } 1267 } 1268 1269 static void jbd2_stats_proc_exit(journal_t *journal) 1270 { 1271 remove_proc_entry("info", journal->j_proc_entry); 1272 remove_proc_entry(journal->j_devname, proc_jbd2_stats); 1273 } 1274 1275 /* Minimum size of descriptor tag */ 1276 static int jbd2_min_tag_size(void) 1277 { 1278 /* 1279 * Tag with 32-bit block numbers does not use last four bytes of the 1280 * structure 1281 */ 1282 return sizeof(journal_block_tag_t) - 4; 1283 } 1284 1285 /** 1286 * jbd2_journal_shrink_scan() 1287 * @shrink: shrinker to work on 1288 * @sc: reclaim request to process 1289 * 1290 * Scan the checkpointed buffer on the checkpoint list and release the 1291 * journal_head. 1292 */ 1293 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink, 1294 struct shrink_control *sc) 1295 { 1296 journal_t *journal = container_of(shrink, journal_t, j_shrinker); 1297 unsigned long nr_to_scan = sc->nr_to_scan; 1298 unsigned long nr_shrunk; 1299 unsigned long count; 1300 1301 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1302 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count); 1303 1304 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan); 1305 1306 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1307 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count); 1308 1309 return nr_shrunk; 1310 } 1311 1312 /** 1313 * jbd2_journal_shrink_count() 1314 * @shrink: shrinker to work on 1315 * @sc: reclaim request to process 1316 * 1317 * Count the number of checkpoint buffers on the checkpoint list. 1318 */ 1319 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink, 1320 struct shrink_control *sc) 1321 { 1322 journal_t *journal = container_of(shrink, journal_t, j_shrinker); 1323 unsigned long count; 1324 1325 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1326 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count); 1327 1328 return count; 1329 } 1330 1331 /* 1332 * Management for journal control blocks: functions to create and 1333 * destroy journal_t structures, and to initialise and read existing 1334 * journal blocks from disk. */ 1335 1336 /* First: create and setup a journal_t object in memory. We initialise 1337 * very few fields yet: that has to wait until we have created the 1338 * journal structures from from scratch, or loaded them from disk. */ 1339 1340 static journal_t *journal_init_common(struct block_device *bdev, 1341 struct block_device *fs_dev, 1342 unsigned long long start, int len, int blocksize) 1343 { 1344 static struct lock_class_key jbd2_trans_commit_key; 1345 journal_t *journal; 1346 int err; 1347 struct buffer_head *bh; 1348 int n; 1349 1350 journal = kzalloc(sizeof(*journal), GFP_KERNEL); 1351 if (!journal) 1352 return NULL; 1353 1354 init_waitqueue_head(&journal->j_wait_transaction_locked); 1355 init_waitqueue_head(&journal->j_wait_done_commit); 1356 init_waitqueue_head(&journal->j_wait_commit); 1357 init_waitqueue_head(&journal->j_wait_updates); 1358 init_waitqueue_head(&journal->j_wait_reserved); 1359 init_waitqueue_head(&journal->j_fc_wait); 1360 mutex_init(&journal->j_abort_mutex); 1361 mutex_init(&journal->j_barrier); 1362 mutex_init(&journal->j_checkpoint_mutex); 1363 spin_lock_init(&journal->j_revoke_lock); 1364 spin_lock_init(&journal->j_list_lock); 1365 rwlock_init(&journal->j_state_lock); 1366 1367 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); 1368 journal->j_min_batch_time = 0; 1369 journal->j_max_batch_time = 15000; /* 15ms */ 1370 atomic_set(&journal->j_reserved_credits, 0); 1371 1372 /* The journal is marked for error until we succeed with recovery! */ 1373 journal->j_flags = JBD2_ABORT; 1374 1375 /* Set up a default-sized revoke table for the new mount. */ 1376 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); 1377 if (err) 1378 goto err_cleanup; 1379 1380 spin_lock_init(&journal->j_history_lock); 1381 1382 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle", 1383 &jbd2_trans_commit_key, 0); 1384 1385 /* journal descriptor can store up to n blocks -bzzz */ 1386 journal->j_blocksize = blocksize; 1387 journal->j_dev = bdev; 1388 journal->j_fs_dev = fs_dev; 1389 journal->j_blk_offset = start; 1390 journal->j_total_len = len; 1391 /* We need enough buffers to write out full descriptor block. */ 1392 n = journal->j_blocksize / jbd2_min_tag_size(); 1393 journal->j_wbufsize = n; 1394 journal->j_fc_wbuf = NULL; 1395 journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *), 1396 GFP_KERNEL); 1397 if (!journal->j_wbuf) 1398 goto err_cleanup; 1399 1400 bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize); 1401 if (!bh) { 1402 pr_err("%s: Cannot get buffer for journal superblock\n", 1403 __func__); 1404 goto err_cleanup; 1405 } 1406 journal->j_sb_buffer = bh; 1407 journal->j_superblock = (journal_superblock_t *)bh->b_data; 1408 1409 journal->j_shrink_transaction = NULL; 1410 journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan; 1411 journal->j_shrinker.count_objects = jbd2_journal_shrink_count; 1412 journal->j_shrinker.seeks = DEFAULT_SEEKS; 1413 journal->j_shrinker.batch = journal->j_max_transaction_buffers; 1414 1415 if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL)) 1416 goto err_cleanup; 1417 1418 if (register_shrinker(&journal->j_shrinker, "jbd2-journal:(%u:%u)", 1419 MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev))) { 1420 percpu_counter_destroy(&journal->j_checkpoint_jh_count); 1421 goto err_cleanup; 1422 } 1423 return journal; 1424 1425 err_cleanup: 1426 brelse(journal->j_sb_buffer); 1427 kfree(journal->j_wbuf); 1428 jbd2_journal_destroy_revoke(journal); 1429 kfree(journal); 1430 return NULL; 1431 } 1432 1433 /* jbd2_journal_init_dev and jbd2_journal_init_inode: 1434 * 1435 * Create a journal structure assigned some fixed set of disk blocks to 1436 * the journal. We don't actually touch those disk blocks yet, but we 1437 * need to set up all of the mapping information to tell the journaling 1438 * system where the journal blocks are. 1439 * 1440 */ 1441 1442 /** 1443 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure 1444 * @bdev: Block device on which to create the journal 1445 * @fs_dev: Device which hold journalled filesystem for this journal. 1446 * @start: Block nr Start of journal. 1447 * @len: Length of the journal in blocks. 1448 * @blocksize: blocksize of journalling device 1449 * 1450 * Returns: a newly created journal_t * 1451 * 1452 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous 1453 * range of blocks on an arbitrary block device. 1454 * 1455 */ 1456 journal_t *jbd2_journal_init_dev(struct block_device *bdev, 1457 struct block_device *fs_dev, 1458 unsigned long long start, int len, int blocksize) 1459 { 1460 journal_t *journal; 1461 1462 journal = journal_init_common(bdev, fs_dev, start, len, blocksize); 1463 if (!journal) 1464 return NULL; 1465 1466 snprintf(journal->j_devname, sizeof(journal->j_devname), 1467 "%pg", journal->j_dev); 1468 strreplace(journal->j_devname, '/', '!'); 1469 jbd2_stats_proc_init(journal); 1470 1471 return journal; 1472 } 1473 1474 /** 1475 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. 1476 * @inode: An inode to create the journal in 1477 * 1478 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as 1479 * the journal. The inode must exist already, must support bmap() and 1480 * must have all data blocks preallocated. 1481 */ 1482 journal_t *jbd2_journal_init_inode(struct inode *inode) 1483 { 1484 journal_t *journal; 1485 sector_t blocknr; 1486 char *p; 1487 int err = 0; 1488 1489 blocknr = 0; 1490 err = bmap(inode, &blocknr); 1491 1492 if (err || !blocknr) { 1493 pr_err("%s: Cannot locate journal superblock\n", 1494 __func__); 1495 return NULL; 1496 } 1497 1498 jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n", 1499 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size, 1500 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); 1501 1502 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev, 1503 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits, 1504 inode->i_sb->s_blocksize); 1505 if (!journal) 1506 return NULL; 1507 1508 journal->j_inode = inode; 1509 snprintf(journal->j_devname, sizeof(journal->j_devname), 1510 "%pg", journal->j_dev); 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 jbd2_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, blk_opf_t 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_flags: Flags for the journal sb write operation 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, 1669 blk_opf_t write_flags) 1670 { 1671 journal_superblock_t *sb = journal->j_superblock; 1672 int ret; 1673 1674 if (is_journal_aborted(journal)) 1675 return -EIO; 1676 if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) { 1677 jbd2_journal_abort(journal, -EIO); 1678 return -EIO; 1679 } 1680 1681 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1682 jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", 1683 tail_block, tail_tid); 1684 1685 lock_buffer(journal->j_sb_buffer); 1686 sb->s_sequence = cpu_to_be32(tail_tid); 1687 sb->s_start = cpu_to_be32(tail_block); 1688 1689 ret = jbd2_write_superblock(journal, write_flags); 1690 if (ret) 1691 goto out; 1692 1693 /* Log is no longer empty */ 1694 write_lock(&journal->j_state_lock); 1695 WARN_ON(!sb->s_sequence); 1696 journal->j_flags &= ~JBD2_FLUSHED; 1697 write_unlock(&journal->j_state_lock); 1698 1699 out: 1700 return ret; 1701 } 1702 1703 /** 1704 * jbd2_mark_journal_empty() - Mark on disk journal as empty. 1705 * @journal: The journal to update. 1706 * @write_flags: Flags for the journal sb write operation 1707 * 1708 * Update a journal's dynamic superblock fields to show that journal is empty. 1709 * Write updated superblock to disk waiting for IO to complete. 1710 */ 1711 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags) 1712 { 1713 journal_superblock_t *sb = journal->j_superblock; 1714 bool had_fast_commit = false; 1715 1716 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1717 lock_buffer(journal->j_sb_buffer); 1718 if (sb->s_start == 0) { /* Is it already empty? */ 1719 unlock_buffer(journal->j_sb_buffer); 1720 return; 1721 } 1722 1723 jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n", 1724 journal->j_tail_sequence); 1725 1726 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); 1727 sb->s_start = cpu_to_be32(0); 1728 if (jbd2_has_feature_fast_commit(journal)) { 1729 /* 1730 * When journal is clean, no need to commit fast commit flag and 1731 * make file system incompatible with older kernels. 1732 */ 1733 jbd2_clear_feature_fast_commit(journal); 1734 had_fast_commit = true; 1735 } 1736 1737 jbd2_write_superblock(journal, write_flags); 1738 1739 if (had_fast_commit) 1740 jbd2_set_feature_fast_commit(journal); 1741 1742 /* Log is no longer empty */ 1743 write_lock(&journal->j_state_lock); 1744 journal->j_flags |= JBD2_FLUSHED; 1745 write_unlock(&journal->j_state_lock); 1746 } 1747 1748 /** 1749 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock) 1750 * @journal: The journal to erase. 1751 * @flags: A discard/zeroout request is sent for each physically contigous 1752 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or 1753 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation 1754 * to perform. 1755 * 1756 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes 1757 * will be explicitly written if no hardware offload is available, see 1758 * blkdev_issue_zeroout for more details. 1759 */ 1760 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags) 1761 { 1762 int err = 0; 1763 unsigned long block, log_offset; /* logical */ 1764 unsigned long long phys_block, block_start, block_stop; /* physical */ 1765 loff_t byte_start, byte_stop, byte_count; 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 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && 1774 !bdev_max_discard_sectors(journal->j_dev)) 1775 return -EOPNOTSUPP; 1776 1777 /* 1778 * lookup block mapping and issue discard/zeroout for each 1779 * contiguous region 1780 */ 1781 log_offset = be32_to_cpu(journal->j_superblock->s_first); 1782 block_start = ~0ULL; 1783 for (block = log_offset; block < journal->j_total_len; block++) { 1784 err = jbd2_journal_bmap(journal, block, &phys_block); 1785 if (err) { 1786 pr_err("JBD2: bad block at offset %lu", block); 1787 return err; 1788 } 1789 1790 if (block_start == ~0ULL) { 1791 block_start = phys_block; 1792 block_stop = block_start - 1; 1793 } 1794 1795 /* 1796 * last block not contiguous with current block, 1797 * process last contiguous region and return to this block on 1798 * next loop 1799 */ 1800 if (phys_block != block_stop + 1) { 1801 block--; 1802 } else { 1803 block_stop++; 1804 /* 1805 * if this isn't the last block of journal, 1806 * no need to process now because next block may also 1807 * be part of this contiguous region 1808 */ 1809 if (block != journal->j_total_len - 1) 1810 continue; 1811 } 1812 1813 /* 1814 * end of contiguous region or this is last block of journal, 1815 * take care of the region 1816 */ 1817 byte_start = block_start * journal->j_blocksize; 1818 byte_stop = block_stop * journal->j_blocksize; 1819 byte_count = (block_stop - block_start + 1) * 1820 journal->j_blocksize; 1821 1822 truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping, 1823 byte_start, byte_stop); 1824 1825 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) { 1826 err = blkdev_issue_discard(journal->j_dev, 1827 byte_start >> SECTOR_SHIFT, 1828 byte_count >> SECTOR_SHIFT, 1829 GFP_NOFS); 1830 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) { 1831 err = blkdev_issue_zeroout(journal->j_dev, 1832 byte_start >> SECTOR_SHIFT, 1833 byte_count >> SECTOR_SHIFT, 1834 GFP_NOFS, 0); 1835 } 1836 1837 if (unlikely(err != 0)) { 1838 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu", 1839 err, block_start, block_stop); 1840 return err; 1841 } 1842 1843 /* reset start and stop after processing a region */ 1844 block_start = ~0ULL; 1845 } 1846 1847 return blkdev_issue_flush(journal->j_dev); 1848 } 1849 1850 /** 1851 * jbd2_journal_update_sb_errno() - Update error in the journal. 1852 * @journal: The journal to update. 1853 * 1854 * Update a journal's errno. Write updated superblock to disk waiting for IO 1855 * to complete. 1856 */ 1857 void jbd2_journal_update_sb_errno(journal_t *journal) 1858 { 1859 journal_superblock_t *sb = journal->j_superblock; 1860 int errcode; 1861 1862 lock_buffer(journal->j_sb_buffer); 1863 errcode = journal->j_errno; 1864 if (errcode == -ESHUTDOWN) 1865 errcode = 0; 1866 jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode); 1867 sb->s_errno = cpu_to_be32(errcode); 1868 1869 jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA); 1870 } 1871 EXPORT_SYMBOL(jbd2_journal_update_sb_errno); 1872 1873 static int journal_revoke_records_per_block(journal_t *journal) 1874 { 1875 int record_size; 1876 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t); 1877 1878 if (jbd2_has_feature_64bit(journal)) 1879 record_size = 8; 1880 else 1881 record_size = 4; 1882 1883 if (jbd2_journal_has_csum_v2or3(journal)) 1884 space -= sizeof(struct jbd2_journal_block_tail); 1885 return space / record_size; 1886 } 1887 1888 /* 1889 * Read the superblock for a given journal, performing initial 1890 * validation of the format. 1891 */ 1892 static int journal_get_superblock(journal_t *journal) 1893 { 1894 struct buffer_head *bh; 1895 journal_superblock_t *sb; 1896 int err; 1897 1898 bh = journal->j_sb_buffer; 1899 1900 J_ASSERT(bh != NULL); 1901 err = bh_read(bh, 0); 1902 if (err < 0) { 1903 printk(KERN_ERR 1904 "JBD2: IO error reading journal superblock\n"); 1905 goto out; 1906 } 1907 1908 if (buffer_verified(bh)) 1909 return 0; 1910 1911 sb = journal->j_superblock; 1912 1913 err = -EINVAL; 1914 1915 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || 1916 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { 1917 printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); 1918 goto out; 1919 } 1920 1921 switch(be32_to_cpu(sb->s_header.h_blocktype)) { 1922 case JBD2_SUPERBLOCK_V1: 1923 journal->j_format_version = 1; 1924 break; 1925 case JBD2_SUPERBLOCK_V2: 1926 journal->j_format_version = 2; 1927 break; 1928 default: 1929 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); 1930 goto out; 1931 } 1932 1933 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len) 1934 journal->j_total_len = be32_to_cpu(sb->s_maxlen); 1935 else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) { 1936 printk(KERN_WARNING "JBD2: journal file too short\n"); 1937 goto out; 1938 } 1939 1940 if (be32_to_cpu(sb->s_first) == 0 || 1941 be32_to_cpu(sb->s_first) >= journal->j_total_len) { 1942 printk(KERN_WARNING 1943 "JBD2: Invalid start block of journal: %u\n", 1944 be32_to_cpu(sb->s_first)); 1945 goto out; 1946 } 1947 1948 if (jbd2_has_feature_csum2(journal) && 1949 jbd2_has_feature_csum3(journal)) { 1950 /* Can't have checksum v2 and v3 at the same time! */ 1951 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " 1952 "at the same time!\n"); 1953 goto out; 1954 } 1955 1956 if (jbd2_journal_has_csum_v2or3_feature(journal) && 1957 jbd2_has_feature_checksum(journal)) { 1958 /* Can't have checksum v1 and v2 on at the same time! */ 1959 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " 1960 "at the same time!\n"); 1961 goto out; 1962 } 1963 1964 if (!jbd2_verify_csum_type(journal, sb)) { 1965 printk(KERN_ERR "JBD2: Unknown checksum type\n"); 1966 goto out; 1967 } 1968 1969 /* Load the checksum driver */ 1970 if (jbd2_journal_has_csum_v2or3_feature(journal)) { 1971 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 1972 if (IS_ERR(journal->j_chksum_driver)) { 1973 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); 1974 err = PTR_ERR(journal->j_chksum_driver); 1975 journal->j_chksum_driver = NULL; 1976 goto out; 1977 } 1978 } 1979 1980 if (jbd2_journal_has_csum_v2or3(journal)) { 1981 /* Check superblock checksum */ 1982 if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) { 1983 printk(KERN_ERR "JBD2: journal checksum error\n"); 1984 err = -EFSBADCRC; 1985 goto out; 1986 } 1987 1988 /* Precompute checksum seed for all metadata */ 1989 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, 1990 sizeof(sb->s_uuid)); 1991 } 1992 1993 journal->j_revoke_records_per_block = 1994 journal_revoke_records_per_block(journal); 1995 set_buffer_verified(bh); 1996 1997 return 0; 1998 1999 out: 2000 journal_fail_superblock(journal); 2001 return err; 2002 } 2003 2004 /* 2005 * Load the on-disk journal superblock and read the key fields into the 2006 * journal_t. 2007 */ 2008 2009 static int load_superblock(journal_t *journal) 2010 { 2011 int err; 2012 journal_superblock_t *sb; 2013 int num_fc_blocks; 2014 2015 err = journal_get_superblock(journal); 2016 if (err) 2017 return err; 2018 2019 sb = journal->j_superblock; 2020 2021 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); 2022 journal->j_tail = be32_to_cpu(sb->s_start); 2023 journal->j_first = be32_to_cpu(sb->s_first); 2024 journal->j_errno = be32_to_cpu(sb->s_errno); 2025 journal->j_last = be32_to_cpu(sb->s_maxlen); 2026 2027 if (jbd2_has_feature_fast_commit(journal)) { 2028 journal->j_fc_last = be32_to_cpu(sb->s_maxlen); 2029 num_fc_blocks = jbd2_journal_get_num_fc_blks(sb); 2030 if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS) 2031 journal->j_last = journal->j_fc_last - num_fc_blocks; 2032 journal->j_fc_first = journal->j_last + 1; 2033 journal->j_fc_off = 0; 2034 } 2035 2036 return 0; 2037 } 2038 2039 2040 /** 2041 * jbd2_journal_load() - Read journal from disk. 2042 * @journal: Journal to act on. 2043 * 2044 * Given a journal_t structure which tells us which disk blocks contain 2045 * a journal, read the journal from disk to initialise the in-memory 2046 * structures. 2047 */ 2048 int jbd2_journal_load(journal_t *journal) 2049 { 2050 int err; 2051 journal_superblock_t *sb; 2052 2053 err = load_superblock(journal); 2054 if (err) 2055 return err; 2056 2057 sb = journal->j_superblock; 2058 /* If this is a V2 superblock, then we have to check the 2059 * features flags on it. */ 2060 2061 if (journal->j_format_version >= 2) { 2062 if ((sb->s_feature_ro_compat & 2063 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || 2064 (sb->s_feature_incompat & 2065 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { 2066 printk(KERN_WARNING 2067 "JBD2: Unrecognised features on journal\n"); 2068 return -EINVAL; 2069 } 2070 } 2071 2072 /* 2073 * Create a slab for this blocksize 2074 */ 2075 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); 2076 if (err) 2077 return err; 2078 2079 /* Let the recovery code check whether it needs to recover any 2080 * data from the journal. */ 2081 if (jbd2_journal_recover(journal)) 2082 goto recovery_error; 2083 2084 if (journal->j_failed_commit) { 2085 printk(KERN_ERR "JBD2: journal transaction %u on %s " 2086 "is corrupt.\n", journal->j_failed_commit, 2087 journal->j_devname); 2088 return -EFSCORRUPTED; 2089 } 2090 /* 2091 * clear JBD2_ABORT flag initialized in journal_init_common 2092 * here to update log tail information with the newest seq. 2093 */ 2094 journal->j_flags &= ~JBD2_ABORT; 2095 2096 /* OK, we've finished with the dynamic journal bits: 2097 * reinitialise the dynamic contents of the superblock in memory 2098 * and reset them on disk. */ 2099 if (journal_reset(journal)) 2100 goto recovery_error; 2101 2102 journal->j_flags |= JBD2_LOADED; 2103 return 0; 2104 2105 recovery_error: 2106 printk(KERN_WARNING "JBD2: recovery failed\n"); 2107 return -EIO; 2108 } 2109 2110 /** 2111 * jbd2_journal_destroy() - Release a journal_t structure. 2112 * @journal: Journal to act on. 2113 * 2114 * Release a journal_t structure once it is no longer in use by the 2115 * journaled object. 2116 * Return <0 if we couldn't clean up the journal. 2117 */ 2118 int jbd2_journal_destroy(journal_t *journal) 2119 { 2120 int err = 0; 2121 2122 /* Wait for the commit thread to wake up and die. */ 2123 journal_kill_thread(journal); 2124 2125 /* Force a final log commit */ 2126 if (journal->j_running_transaction) 2127 jbd2_journal_commit_transaction(journal); 2128 2129 /* Force any old transactions to disk */ 2130 2131 /* Totally anal locking here... */ 2132 spin_lock(&journal->j_list_lock); 2133 while (journal->j_checkpoint_transactions != NULL) { 2134 spin_unlock(&journal->j_list_lock); 2135 mutex_lock_io(&journal->j_checkpoint_mutex); 2136 err = jbd2_log_do_checkpoint(journal); 2137 mutex_unlock(&journal->j_checkpoint_mutex); 2138 /* 2139 * If checkpointing failed, just free the buffers to avoid 2140 * looping forever 2141 */ 2142 if (err) { 2143 jbd2_journal_destroy_checkpoint(journal); 2144 spin_lock(&journal->j_list_lock); 2145 break; 2146 } 2147 spin_lock(&journal->j_list_lock); 2148 } 2149 2150 J_ASSERT(journal->j_running_transaction == NULL); 2151 J_ASSERT(journal->j_committing_transaction == NULL); 2152 J_ASSERT(journal->j_checkpoint_transactions == NULL); 2153 spin_unlock(&journal->j_list_lock); 2154 2155 /* 2156 * OK, all checkpoint transactions have been checked, now check the 2157 * write out io error flag and abort the journal if some buffer failed 2158 * to write back to the original location, otherwise the filesystem 2159 * may become inconsistent. 2160 */ 2161 if (!is_journal_aborted(journal) && 2162 test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) 2163 jbd2_journal_abort(journal, -EIO); 2164 2165 if (journal->j_sb_buffer) { 2166 if (!is_journal_aborted(journal)) { 2167 mutex_lock_io(&journal->j_checkpoint_mutex); 2168 2169 write_lock(&journal->j_state_lock); 2170 journal->j_tail_sequence = 2171 ++journal->j_transaction_sequence; 2172 write_unlock(&journal->j_state_lock); 2173 2174 jbd2_mark_journal_empty(journal, 2175 REQ_SYNC | REQ_PREFLUSH | REQ_FUA); 2176 mutex_unlock(&journal->j_checkpoint_mutex); 2177 } else 2178 err = -EIO; 2179 brelse(journal->j_sb_buffer); 2180 } 2181 2182 if (journal->j_shrinker.flags & SHRINKER_REGISTERED) { 2183 percpu_counter_destroy(&journal->j_checkpoint_jh_count); 2184 unregister_shrinker(&journal->j_shrinker); 2185 } 2186 if (journal->j_proc_entry) 2187 jbd2_stats_proc_exit(journal); 2188 iput(journal->j_inode); 2189 if (journal->j_revoke) 2190 jbd2_journal_destroy_revoke(journal); 2191 if (journal->j_chksum_driver) 2192 crypto_free_shash(journal->j_chksum_driver); 2193 kfree(journal->j_fc_wbuf); 2194 kfree(journal->j_wbuf); 2195 kfree(journal); 2196 2197 return err; 2198 } 2199 2200 2201 /** 2202 * jbd2_journal_check_used_features() - Check if features specified are used. 2203 * @journal: Journal to check. 2204 * @compat: bitmask of compatible features 2205 * @ro: bitmask of features that force read-only mount 2206 * @incompat: bitmask of incompatible features 2207 * 2208 * Check whether the journal uses all of a given set of 2209 * features. Return true (non-zero) if it does. 2210 **/ 2211 2212 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat, 2213 unsigned long ro, unsigned long incompat) 2214 { 2215 journal_superblock_t *sb; 2216 2217 if (!compat && !ro && !incompat) 2218 return 1; 2219 /* Load journal superblock if it is not loaded yet. */ 2220 if (journal->j_format_version == 0 && 2221 journal_get_superblock(journal) != 0) 2222 return 0; 2223 if (journal->j_format_version == 1) 2224 return 0; 2225 2226 sb = journal->j_superblock; 2227 2228 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && 2229 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && 2230 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) 2231 return 1; 2232 2233 return 0; 2234 } 2235 2236 /** 2237 * jbd2_journal_check_available_features() - Check feature set in journalling layer 2238 * @journal: Journal to check. 2239 * @compat: bitmask of compatible features 2240 * @ro: bitmask of features that force read-only mount 2241 * @incompat: bitmask of incompatible features 2242 * 2243 * Check whether the journaling code supports the use of 2244 * all of a given set of features on this journal. Return true 2245 * (non-zero) if it can. */ 2246 2247 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat, 2248 unsigned long ro, unsigned long incompat) 2249 { 2250 if (!compat && !ro && !incompat) 2251 return 1; 2252 2253 /* We can support any known requested features iff the 2254 * superblock is in version 2. Otherwise we fail to support any 2255 * extended sb features. */ 2256 2257 if (journal->j_format_version != 2) 2258 return 0; 2259 2260 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && 2261 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && 2262 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) 2263 return 1; 2264 2265 return 0; 2266 } 2267 2268 static int 2269 jbd2_journal_initialize_fast_commit(journal_t *journal) 2270 { 2271 journal_superblock_t *sb = journal->j_superblock; 2272 unsigned long long num_fc_blks; 2273 2274 num_fc_blks = jbd2_journal_get_num_fc_blks(sb); 2275 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS) 2276 return -ENOSPC; 2277 2278 /* Are we called twice? */ 2279 WARN_ON(journal->j_fc_wbuf != NULL); 2280 journal->j_fc_wbuf = kmalloc_array(num_fc_blks, 2281 sizeof(struct buffer_head *), GFP_KERNEL); 2282 if (!journal->j_fc_wbuf) 2283 return -ENOMEM; 2284 2285 journal->j_fc_wbufsize = num_fc_blks; 2286 journal->j_fc_last = journal->j_last; 2287 journal->j_last = journal->j_fc_last - num_fc_blks; 2288 journal->j_fc_first = journal->j_last + 1; 2289 journal->j_fc_off = 0; 2290 journal->j_free = journal->j_last - journal->j_first; 2291 journal->j_max_transaction_buffers = 2292 jbd2_journal_get_max_txn_bufs(journal); 2293 2294 return 0; 2295 } 2296 2297 /** 2298 * jbd2_journal_set_features() - Mark a given journal feature in the superblock 2299 * @journal: Journal to act on. 2300 * @compat: bitmask of compatible features 2301 * @ro: bitmask of features that force read-only mount 2302 * @incompat: bitmask of incompatible features 2303 * 2304 * Mark a given journal feature as present on the 2305 * superblock. Returns true if the requested features could be set. 2306 * 2307 */ 2308 2309 int jbd2_journal_set_features(journal_t *journal, unsigned long compat, 2310 unsigned long ro, unsigned long incompat) 2311 { 2312 #define INCOMPAT_FEATURE_ON(f) \ 2313 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) 2314 #define COMPAT_FEATURE_ON(f) \ 2315 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) 2316 journal_superblock_t *sb; 2317 2318 if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) 2319 return 1; 2320 2321 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) 2322 return 0; 2323 2324 /* If enabling v2 checksums, turn on v3 instead */ 2325 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { 2326 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; 2327 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; 2328 } 2329 2330 /* Asking for checksumming v3 and v1? Only give them v3. */ 2331 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && 2332 compat & JBD2_FEATURE_COMPAT_CHECKSUM) 2333 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; 2334 2335 jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", 2336 compat, ro, incompat); 2337 2338 sb = journal->j_superblock; 2339 2340 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) { 2341 if (jbd2_journal_initialize_fast_commit(journal)) { 2342 pr_err("JBD2: Cannot enable fast commits.\n"); 2343 return 0; 2344 } 2345 } 2346 2347 /* Load the checksum driver if necessary */ 2348 if ((journal->j_chksum_driver == NULL) && 2349 INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 2350 journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); 2351 if (IS_ERR(journal->j_chksum_driver)) { 2352 printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); 2353 journal->j_chksum_driver = NULL; 2354 return 0; 2355 } 2356 /* Precompute checksum seed for all metadata */ 2357 journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, 2358 sizeof(sb->s_uuid)); 2359 } 2360 2361 lock_buffer(journal->j_sb_buffer); 2362 2363 /* If enabling v3 checksums, update superblock */ 2364 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 2365 sb->s_checksum_type = JBD2_CRC32C_CHKSUM; 2366 sb->s_feature_compat &= 2367 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); 2368 } 2369 2370 /* If enabling v1 checksums, downgrade superblock */ 2371 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) 2372 sb->s_feature_incompat &= 2373 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | 2374 JBD2_FEATURE_INCOMPAT_CSUM_V3); 2375 2376 sb->s_feature_compat |= cpu_to_be32(compat); 2377 sb->s_feature_ro_compat |= cpu_to_be32(ro); 2378 sb->s_feature_incompat |= cpu_to_be32(incompat); 2379 unlock_buffer(journal->j_sb_buffer); 2380 journal->j_revoke_records_per_block = 2381 journal_revoke_records_per_block(journal); 2382 2383 return 1; 2384 #undef COMPAT_FEATURE_ON 2385 #undef INCOMPAT_FEATURE_ON 2386 } 2387 2388 /* 2389 * jbd2_journal_clear_features() - Clear a given journal feature in the 2390 * superblock 2391 * @journal: Journal to act on. 2392 * @compat: bitmask of compatible features 2393 * @ro: bitmask of features that force read-only mount 2394 * @incompat: bitmask of incompatible features 2395 * 2396 * Clear a given journal feature as present on the 2397 * superblock. 2398 */ 2399 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, 2400 unsigned long ro, unsigned long incompat) 2401 { 2402 journal_superblock_t *sb; 2403 2404 jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", 2405 compat, ro, incompat); 2406 2407 sb = journal->j_superblock; 2408 2409 sb->s_feature_compat &= ~cpu_to_be32(compat); 2410 sb->s_feature_ro_compat &= ~cpu_to_be32(ro); 2411 sb->s_feature_incompat &= ~cpu_to_be32(incompat); 2412 journal->j_revoke_records_per_block = 2413 journal_revoke_records_per_block(journal); 2414 } 2415 EXPORT_SYMBOL(jbd2_journal_clear_features); 2416 2417 /** 2418 * jbd2_journal_flush() - Flush journal 2419 * @journal: Journal to act on. 2420 * @flags: optional operation on the journal blocks after the flush (see below) 2421 * 2422 * Flush all data for a given journal to disk and empty the journal. 2423 * Filesystems can use this when remounting readonly to ensure that 2424 * recovery does not need to happen on remount. Optionally, a discard or zeroout 2425 * can be issued on the journal blocks after flushing. 2426 * 2427 * flags: 2428 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks 2429 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks 2430 */ 2431 int jbd2_journal_flush(journal_t *journal, unsigned int flags) 2432 { 2433 int err = 0; 2434 transaction_t *transaction = NULL; 2435 2436 write_lock(&journal->j_state_lock); 2437 2438 /* Force everything buffered to the log... */ 2439 if (journal->j_running_transaction) { 2440 transaction = journal->j_running_transaction; 2441 __jbd2_log_start_commit(journal, transaction->t_tid); 2442 } else if (journal->j_committing_transaction) 2443 transaction = journal->j_committing_transaction; 2444 2445 /* Wait for the log commit to complete... */ 2446 if (transaction) { 2447 tid_t tid = transaction->t_tid; 2448 2449 write_unlock(&journal->j_state_lock); 2450 jbd2_log_wait_commit(journal, tid); 2451 } else { 2452 write_unlock(&journal->j_state_lock); 2453 } 2454 2455 /* ...and flush everything in the log out to disk. */ 2456 spin_lock(&journal->j_list_lock); 2457 while (!err && journal->j_checkpoint_transactions != NULL) { 2458 spin_unlock(&journal->j_list_lock); 2459 mutex_lock_io(&journal->j_checkpoint_mutex); 2460 err = jbd2_log_do_checkpoint(journal); 2461 mutex_unlock(&journal->j_checkpoint_mutex); 2462 spin_lock(&journal->j_list_lock); 2463 } 2464 spin_unlock(&journal->j_list_lock); 2465 2466 if (is_journal_aborted(journal)) 2467 return -EIO; 2468 2469 mutex_lock_io(&journal->j_checkpoint_mutex); 2470 if (!err) { 2471 err = jbd2_cleanup_journal_tail(journal); 2472 if (err < 0) { 2473 mutex_unlock(&journal->j_checkpoint_mutex); 2474 goto out; 2475 } 2476 err = 0; 2477 } 2478 2479 /* Finally, mark the journal as really needing no recovery. 2480 * This sets s_start==0 in the underlying superblock, which is 2481 * the magic code for a fully-recovered superblock. Any future 2482 * commits of data to the journal will restore the current 2483 * s_start value. */ 2484 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); 2485 2486 if (flags) 2487 err = __jbd2_journal_erase(journal, flags); 2488 2489 mutex_unlock(&journal->j_checkpoint_mutex); 2490 write_lock(&journal->j_state_lock); 2491 J_ASSERT(!journal->j_running_transaction); 2492 J_ASSERT(!journal->j_committing_transaction); 2493 J_ASSERT(!journal->j_checkpoint_transactions); 2494 J_ASSERT(journal->j_head == journal->j_tail); 2495 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); 2496 write_unlock(&journal->j_state_lock); 2497 out: 2498 return err; 2499 } 2500 2501 /** 2502 * jbd2_journal_wipe() - Wipe journal contents 2503 * @journal: Journal to act on. 2504 * @write: flag (see below) 2505 * 2506 * Wipe out all of the contents of a journal, safely. This will produce 2507 * a warning if the journal contains any valid recovery information. 2508 * Must be called between journal_init_*() and jbd2_journal_load(). 2509 * 2510 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise 2511 * we merely suppress recovery. 2512 */ 2513 2514 int jbd2_journal_wipe(journal_t *journal, int write) 2515 { 2516 int err = 0; 2517 2518 J_ASSERT (!(journal->j_flags & JBD2_LOADED)); 2519 2520 err = load_superblock(journal); 2521 if (err) 2522 return err; 2523 2524 if (!journal->j_tail) 2525 goto no_recovery; 2526 2527 printk(KERN_WARNING "JBD2: %s recovery information on journal\n", 2528 write ? "Clearing" : "Ignoring"); 2529 2530 err = jbd2_journal_skip_recovery(journal); 2531 if (write) { 2532 /* Lock to make assertions happy... */ 2533 mutex_lock_io(&journal->j_checkpoint_mutex); 2534 jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); 2535 mutex_unlock(&journal->j_checkpoint_mutex); 2536 } 2537 2538 no_recovery: 2539 return err; 2540 } 2541 2542 /** 2543 * jbd2_journal_abort () - Shutdown the journal immediately. 2544 * @journal: the journal to shutdown. 2545 * @errno: an error number to record in the journal indicating 2546 * the reason for the shutdown. 2547 * 2548 * Perform a complete, immediate shutdown of the ENTIRE 2549 * journal (not of a single transaction). This operation cannot be 2550 * undone without closing and reopening the journal. 2551 * 2552 * The jbd2_journal_abort function is intended to support higher level error 2553 * recovery mechanisms such as the ext2/ext3 remount-readonly error 2554 * mode. 2555 * 2556 * Journal abort has very specific semantics. Any existing dirty, 2557 * unjournaled buffers in the main filesystem will still be written to 2558 * disk by bdflush, but the journaling mechanism will be suspended 2559 * immediately and no further transaction commits will be honoured. 2560 * 2561 * Any dirty, journaled buffers will be written back to disk without 2562 * hitting the journal. Atomicity cannot be guaranteed on an aborted 2563 * filesystem, but we _do_ attempt to leave as much data as possible 2564 * behind for fsck to use for cleanup. 2565 * 2566 * Any attempt to get a new transaction handle on a journal which is in 2567 * ABORT state will just result in an -EROFS error return. A 2568 * jbd2_journal_stop on an existing handle will return -EIO if we have 2569 * entered abort state during the update. 2570 * 2571 * Recursive transactions are not disturbed by journal abort until the 2572 * final jbd2_journal_stop, which will receive the -EIO error. 2573 * 2574 * Finally, the jbd2_journal_abort call allows the caller to supply an errno 2575 * which will be recorded (if possible) in the journal superblock. This 2576 * allows a client to record failure conditions in the middle of a 2577 * transaction without having to complete the transaction to record the 2578 * failure to disk. ext3_error, for example, now uses this 2579 * functionality. 2580 * 2581 */ 2582 2583 void jbd2_journal_abort(journal_t *journal, int errno) 2584 { 2585 transaction_t *transaction; 2586 2587 /* 2588 * Lock the aborting procedure until everything is done, this avoid 2589 * races between filesystem's error handling flow (e.g. ext4_abort()), 2590 * ensure panic after the error info is written into journal's 2591 * superblock. 2592 */ 2593 mutex_lock(&journal->j_abort_mutex); 2594 /* 2595 * ESHUTDOWN always takes precedence because a file system check 2596 * caused by any other journal abort error is not required after 2597 * a shutdown triggered. 2598 */ 2599 write_lock(&journal->j_state_lock); 2600 if (journal->j_flags & JBD2_ABORT) { 2601 int old_errno = journal->j_errno; 2602 2603 write_unlock(&journal->j_state_lock); 2604 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) { 2605 journal->j_errno = errno; 2606 jbd2_journal_update_sb_errno(journal); 2607 } 2608 mutex_unlock(&journal->j_abort_mutex); 2609 return; 2610 } 2611 2612 /* 2613 * Mark the abort as occurred and start current running transaction 2614 * to release all journaled buffer. 2615 */ 2616 pr_err("Aborting journal on device %s.\n", journal->j_devname); 2617 2618 journal->j_flags |= JBD2_ABORT; 2619 journal->j_errno = errno; 2620 transaction = journal->j_running_transaction; 2621 if (transaction) 2622 __jbd2_log_start_commit(journal, transaction->t_tid); 2623 write_unlock(&journal->j_state_lock); 2624 2625 /* 2626 * Record errno to the journal super block, so that fsck and jbd2 2627 * layer could realise that a filesystem check is needed. 2628 */ 2629 jbd2_journal_update_sb_errno(journal); 2630 mutex_unlock(&journal->j_abort_mutex); 2631 } 2632 2633 /** 2634 * jbd2_journal_errno() - returns the journal's error state. 2635 * @journal: journal to examine. 2636 * 2637 * This is the errno number set with jbd2_journal_abort(), the last 2638 * time the journal was mounted - if the journal was stopped 2639 * without calling abort this will be 0. 2640 * 2641 * If the journal has been aborted on this mount time -EROFS will 2642 * be returned. 2643 */ 2644 int jbd2_journal_errno(journal_t *journal) 2645 { 2646 int err; 2647 2648 read_lock(&journal->j_state_lock); 2649 if (journal->j_flags & JBD2_ABORT) 2650 err = -EROFS; 2651 else 2652 err = journal->j_errno; 2653 read_unlock(&journal->j_state_lock); 2654 return err; 2655 } 2656 2657 /** 2658 * jbd2_journal_clear_err() - clears the journal's error state 2659 * @journal: journal to act on. 2660 * 2661 * An error must be cleared or acked to take a FS out of readonly 2662 * mode. 2663 */ 2664 int jbd2_journal_clear_err(journal_t *journal) 2665 { 2666 int err = 0; 2667 2668 write_lock(&journal->j_state_lock); 2669 if (journal->j_flags & JBD2_ABORT) 2670 err = -EROFS; 2671 else 2672 journal->j_errno = 0; 2673 write_unlock(&journal->j_state_lock); 2674 return err; 2675 } 2676 2677 /** 2678 * jbd2_journal_ack_err() - Ack journal err. 2679 * @journal: journal to act on. 2680 * 2681 * An error must be cleared or acked to take a FS out of readonly 2682 * mode. 2683 */ 2684 void jbd2_journal_ack_err(journal_t *journal) 2685 { 2686 write_lock(&journal->j_state_lock); 2687 if (journal->j_errno) 2688 journal->j_flags |= JBD2_ACK_ERR; 2689 write_unlock(&journal->j_state_lock); 2690 } 2691 2692 int jbd2_journal_blocks_per_page(struct inode *inode) 2693 { 2694 return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); 2695 } 2696 2697 /* 2698 * helper functions to deal with 32 or 64bit block numbers. 2699 */ 2700 size_t journal_tag_bytes(journal_t *journal) 2701 { 2702 size_t sz; 2703 2704 if (jbd2_has_feature_csum3(journal)) 2705 return sizeof(journal_block_tag3_t); 2706 2707 sz = sizeof(journal_block_tag_t); 2708 2709 if (jbd2_has_feature_csum2(journal)) 2710 sz += sizeof(__u16); 2711 2712 if (jbd2_has_feature_64bit(journal)) 2713 return sz; 2714 else 2715 return sz - sizeof(__u32); 2716 } 2717 2718 /* 2719 * JBD memory management 2720 * 2721 * These functions are used to allocate block-sized chunks of memory 2722 * used for making copies of buffer_head data. Very often it will be 2723 * page-sized chunks of data, but sometimes it will be in 2724 * sub-page-size chunks. (For example, 16k pages on Power systems 2725 * with a 4k block file system.) For blocks smaller than a page, we 2726 * use a SLAB allocator. There are slab caches for each block size, 2727 * which are allocated at mount time, if necessary, and we only free 2728 * (all of) the slab caches when/if the jbd2 module is unloaded. For 2729 * this reason we don't need to a mutex to protect access to 2730 * jbd2_slab[] allocating or releasing memory; only in 2731 * jbd2_journal_create_slab(). 2732 */ 2733 #define JBD2_MAX_SLABS 8 2734 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; 2735 2736 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { 2737 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", 2738 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" 2739 }; 2740 2741 2742 static void jbd2_journal_destroy_slabs(void) 2743 { 2744 int i; 2745 2746 for (i = 0; i < JBD2_MAX_SLABS; i++) { 2747 kmem_cache_destroy(jbd2_slab[i]); 2748 jbd2_slab[i] = NULL; 2749 } 2750 } 2751 2752 static int jbd2_journal_create_slab(size_t size) 2753 { 2754 static DEFINE_MUTEX(jbd2_slab_create_mutex); 2755 int i = order_base_2(size) - 10; 2756 size_t slab_size; 2757 2758 if (size == PAGE_SIZE) 2759 return 0; 2760 2761 if (i >= JBD2_MAX_SLABS) 2762 return -EINVAL; 2763 2764 if (unlikely(i < 0)) 2765 i = 0; 2766 mutex_lock(&jbd2_slab_create_mutex); 2767 if (jbd2_slab[i]) { 2768 mutex_unlock(&jbd2_slab_create_mutex); 2769 return 0; /* Already created */ 2770 } 2771 2772 slab_size = 1 << (i+10); 2773 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, 2774 slab_size, 0, NULL); 2775 mutex_unlock(&jbd2_slab_create_mutex); 2776 if (!jbd2_slab[i]) { 2777 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); 2778 return -ENOMEM; 2779 } 2780 return 0; 2781 } 2782 2783 static struct kmem_cache *get_slab(size_t size) 2784 { 2785 int i = order_base_2(size) - 10; 2786 2787 BUG_ON(i >= JBD2_MAX_SLABS); 2788 if (unlikely(i < 0)) 2789 i = 0; 2790 BUG_ON(jbd2_slab[i] == NULL); 2791 return jbd2_slab[i]; 2792 } 2793 2794 void *jbd2_alloc(size_t size, gfp_t flags) 2795 { 2796 void *ptr; 2797 2798 BUG_ON(size & (size-1)); /* Must be a power of 2 */ 2799 2800 if (size < PAGE_SIZE) 2801 ptr = kmem_cache_alloc(get_slab(size), flags); 2802 else 2803 ptr = (void *)__get_free_pages(flags, get_order(size)); 2804 2805 /* Check alignment; SLUB has gotten this wrong in the past, 2806 * and this can lead to user data corruption! */ 2807 BUG_ON(((unsigned long) ptr) & (size-1)); 2808 2809 return ptr; 2810 } 2811 2812 void jbd2_free(void *ptr, size_t size) 2813 { 2814 if (size < PAGE_SIZE) 2815 kmem_cache_free(get_slab(size), ptr); 2816 else 2817 free_pages((unsigned long)ptr, get_order(size)); 2818 }; 2819 2820 /* 2821 * Journal_head storage management 2822 */ 2823 static struct kmem_cache *jbd2_journal_head_cache; 2824 #ifdef CONFIG_JBD2_DEBUG 2825 static atomic_t nr_journal_heads = ATOMIC_INIT(0); 2826 #endif 2827 2828 static int __init jbd2_journal_init_journal_head_cache(void) 2829 { 2830 J_ASSERT(!jbd2_journal_head_cache); 2831 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", 2832 sizeof(struct journal_head), 2833 0, /* offset */ 2834 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU, 2835 NULL); /* ctor */ 2836 if (!jbd2_journal_head_cache) { 2837 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); 2838 return -ENOMEM; 2839 } 2840 return 0; 2841 } 2842 2843 static void jbd2_journal_destroy_journal_head_cache(void) 2844 { 2845 kmem_cache_destroy(jbd2_journal_head_cache); 2846 jbd2_journal_head_cache = NULL; 2847 } 2848 2849 /* 2850 * journal_head splicing and dicing 2851 */ 2852 static struct journal_head *journal_alloc_journal_head(void) 2853 { 2854 struct journal_head *ret; 2855 2856 #ifdef CONFIG_JBD2_DEBUG 2857 atomic_inc(&nr_journal_heads); 2858 #endif 2859 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); 2860 if (!ret) { 2861 jbd2_debug(1, "out of memory for journal_head\n"); 2862 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); 2863 ret = kmem_cache_zalloc(jbd2_journal_head_cache, 2864 GFP_NOFS | __GFP_NOFAIL); 2865 } 2866 if (ret) 2867 spin_lock_init(&ret->b_state_lock); 2868 return ret; 2869 } 2870 2871 static void journal_free_journal_head(struct journal_head *jh) 2872 { 2873 #ifdef CONFIG_JBD2_DEBUG 2874 atomic_dec(&nr_journal_heads); 2875 memset(jh, JBD2_POISON_FREE, sizeof(*jh)); 2876 #endif 2877 kmem_cache_free(jbd2_journal_head_cache, jh); 2878 } 2879 2880 /* 2881 * A journal_head is attached to a buffer_head whenever JBD has an 2882 * interest in the buffer. 2883 * 2884 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit 2885 * is set. This bit is tested in core kernel code where we need to take 2886 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable 2887 * there. 2888 * 2889 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. 2890 * 2891 * When a buffer has its BH_JBD bit set it is immune from being released by 2892 * core kernel code, mainly via ->b_count. 2893 * 2894 * A journal_head is detached from its buffer_head when the journal_head's 2895 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint 2896 * transaction (b_cp_transaction) hold their references to b_jcount. 2897 * 2898 * Various places in the kernel want to attach a journal_head to a buffer_head 2899 * _before_ attaching the journal_head to a transaction. To protect the 2900 * journal_head in this situation, jbd2_journal_add_journal_head elevates the 2901 * journal_head's b_jcount refcount by one. The caller must call 2902 * jbd2_journal_put_journal_head() to undo this. 2903 * 2904 * So the typical usage would be: 2905 * 2906 * (Attach a journal_head if needed. Increments b_jcount) 2907 * struct journal_head *jh = jbd2_journal_add_journal_head(bh); 2908 * ... 2909 * (Get another reference for transaction) 2910 * jbd2_journal_grab_journal_head(bh); 2911 * jh->b_transaction = xxx; 2912 * (Put original reference) 2913 * jbd2_journal_put_journal_head(jh); 2914 */ 2915 2916 /* 2917 * Give a buffer_head a journal_head. 2918 * 2919 * May sleep. 2920 */ 2921 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) 2922 { 2923 struct journal_head *jh; 2924 struct journal_head *new_jh = NULL; 2925 2926 repeat: 2927 if (!buffer_jbd(bh)) 2928 new_jh = journal_alloc_journal_head(); 2929 2930 jbd_lock_bh_journal_head(bh); 2931 if (buffer_jbd(bh)) { 2932 jh = bh2jh(bh); 2933 } else { 2934 J_ASSERT_BH(bh, 2935 (atomic_read(&bh->b_count) > 0) || 2936 (bh->b_page && bh->b_page->mapping)); 2937 2938 if (!new_jh) { 2939 jbd_unlock_bh_journal_head(bh); 2940 goto repeat; 2941 } 2942 2943 jh = new_jh; 2944 new_jh = NULL; /* We consumed it */ 2945 set_buffer_jbd(bh); 2946 bh->b_private = jh; 2947 jh->b_bh = bh; 2948 get_bh(bh); 2949 BUFFER_TRACE(bh, "added journal_head"); 2950 } 2951 jh->b_jcount++; 2952 jbd_unlock_bh_journal_head(bh); 2953 if (new_jh) 2954 journal_free_journal_head(new_jh); 2955 return bh->b_private; 2956 } 2957 2958 /* 2959 * Grab a ref against this buffer_head's journal_head. If it ended up not 2960 * having a journal_head, return NULL 2961 */ 2962 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) 2963 { 2964 struct journal_head *jh = NULL; 2965 2966 jbd_lock_bh_journal_head(bh); 2967 if (buffer_jbd(bh)) { 2968 jh = bh2jh(bh); 2969 jh->b_jcount++; 2970 } 2971 jbd_unlock_bh_journal_head(bh); 2972 return jh; 2973 } 2974 EXPORT_SYMBOL(jbd2_journal_grab_journal_head); 2975 2976 static void __journal_remove_journal_head(struct buffer_head *bh) 2977 { 2978 struct journal_head *jh = bh2jh(bh); 2979 2980 J_ASSERT_JH(jh, jh->b_transaction == NULL); 2981 J_ASSERT_JH(jh, jh->b_next_transaction == NULL); 2982 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); 2983 J_ASSERT_JH(jh, jh->b_jlist == BJ_None); 2984 J_ASSERT_BH(bh, buffer_jbd(bh)); 2985 J_ASSERT_BH(bh, jh2bh(jh) == bh); 2986 BUFFER_TRACE(bh, "remove journal_head"); 2987 2988 /* Unlink before dropping the lock */ 2989 bh->b_private = NULL; 2990 jh->b_bh = NULL; /* debug, really */ 2991 clear_buffer_jbd(bh); 2992 } 2993 2994 static void journal_release_journal_head(struct journal_head *jh, size_t b_size) 2995 { 2996 if (jh->b_frozen_data) { 2997 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); 2998 jbd2_free(jh->b_frozen_data, b_size); 2999 } 3000 if (jh->b_committed_data) { 3001 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); 3002 jbd2_free(jh->b_committed_data, b_size); 3003 } 3004 journal_free_journal_head(jh); 3005 } 3006 3007 /* 3008 * Drop a reference on the passed journal_head. If it fell to zero then 3009 * release the journal_head from the buffer_head. 3010 */ 3011 void jbd2_journal_put_journal_head(struct journal_head *jh) 3012 { 3013 struct buffer_head *bh = jh2bh(jh); 3014 3015 jbd_lock_bh_journal_head(bh); 3016 J_ASSERT_JH(jh, jh->b_jcount > 0); 3017 --jh->b_jcount; 3018 if (!jh->b_jcount) { 3019 __journal_remove_journal_head(bh); 3020 jbd_unlock_bh_journal_head(bh); 3021 journal_release_journal_head(jh, bh->b_size); 3022 __brelse(bh); 3023 } else { 3024 jbd_unlock_bh_journal_head(bh); 3025 } 3026 } 3027 EXPORT_SYMBOL(jbd2_journal_put_journal_head); 3028 3029 /* 3030 * Initialize jbd inode head 3031 */ 3032 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) 3033 { 3034 jinode->i_transaction = NULL; 3035 jinode->i_next_transaction = NULL; 3036 jinode->i_vfs_inode = inode; 3037 jinode->i_flags = 0; 3038 jinode->i_dirty_start = 0; 3039 jinode->i_dirty_end = 0; 3040 INIT_LIST_HEAD(&jinode->i_list); 3041 } 3042 3043 /* 3044 * Function to be called before we start removing inode from memory (i.e., 3045 * clear_inode() is a fine place to be called from). It removes inode from 3046 * transaction's lists. 3047 */ 3048 void jbd2_journal_release_jbd_inode(journal_t *journal, 3049 struct jbd2_inode *jinode) 3050 { 3051 if (!journal) 3052 return; 3053 restart: 3054 spin_lock(&journal->j_list_lock); 3055 /* Is commit writing out inode - we have to wait */ 3056 if (jinode->i_flags & JI_COMMIT_RUNNING) { 3057 wait_queue_head_t *wq; 3058 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); 3059 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); 3060 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 3061 spin_unlock(&journal->j_list_lock); 3062 schedule(); 3063 finish_wait(wq, &wait.wq_entry); 3064 goto restart; 3065 } 3066 3067 if (jinode->i_transaction) { 3068 list_del(&jinode->i_list); 3069 jinode->i_transaction = NULL; 3070 } 3071 spin_unlock(&journal->j_list_lock); 3072 } 3073 3074 3075 #ifdef CONFIG_PROC_FS 3076 3077 #define JBD2_STATS_PROC_NAME "fs/jbd2" 3078 3079 static void __init jbd2_create_jbd_stats_proc_entry(void) 3080 { 3081 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); 3082 } 3083 3084 static void __exit jbd2_remove_jbd_stats_proc_entry(void) 3085 { 3086 if (proc_jbd2_stats) 3087 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); 3088 } 3089 3090 #else 3091 3092 #define jbd2_create_jbd_stats_proc_entry() do {} while (0) 3093 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) 3094 3095 #endif 3096 3097 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; 3098 3099 static int __init jbd2_journal_init_inode_cache(void) 3100 { 3101 J_ASSERT(!jbd2_inode_cache); 3102 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); 3103 if (!jbd2_inode_cache) { 3104 pr_emerg("JBD2: failed to create inode cache\n"); 3105 return -ENOMEM; 3106 } 3107 return 0; 3108 } 3109 3110 static int __init jbd2_journal_init_handle_cache(void) 3111 { 3112 J_ASSERT(!jbd2_handle_cache); 3113 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); 3114 if (!jbd2_handle_cache) { 3115 printk(KERN_EMERG "JBD2: failed to create handle cache\n"); 3116 return -ENOMEM; 3117 } 3118 return 0; 3119 } 3120 3121 static void jbd2_journal_destroy_inode_cache(void) 3122 { 3123 kmem_cache_destroy(jbd2_inode_cache); 3124 jbd2_inode_cache = NULL; 3125 } 3126 3127 static void jbd2_journal_destroy_handle_cache(void) 3128 { 3129 kmem_cache_destroy(jbd2_handle_cache); 3130 jbd2_handle_cache = NULL; 3131 } 3132 3133 /* 3134 * Module startup and shutdown 3135 */ 3136 3137 static int __init journal_init_caches(void) 3138 { 3139 int ret; 3140 3141 ret = jbd2_journal_init_revoke_record_cache(); 3142 if (ret == 0) 3143 ret = jbd2_journal_init_revoke_table_cache(); 3144 if (ret == 0) 3145 ret = jbd2_journal_init_journal_head_cache(); 3146 if (ret == 0) 3147 ret = jbd2_journal_init_handle_cache(); 3148 if (ret == 0) 3149 ret = jbd2_journal_init_inode_cache(); 3150 if (ret == 0) 3151 ret = jbd2_journal_init_transaction_cache(); 3152 return ret; 3153 } 3154 3155 static void jbd2_journal_destroy_caches(void) 3156 { 3157 jbd2_journal_destroy_revoke_record_cache(); 3158 jbd2_journal_destroy_revoke_table_cache(); 3159 jbd2_journal_destroy_journal_head_cache(); 3160 jbd2_journal_destroy_handle_cache(); 3161 jbd2_journal_destroy_inode_cache(); 3162 jbd2_journal_destroy_transaction_cache(); 3163 jbd2_journal_destroy_slabs(); 3164 } 3165 3166 static int __init journal_init(void) 3167 { 3168 int ret; 3169 3170 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); 3171 3172 ret = journal_init_caches(); 3173 if (ret == 0) { 3174 jbd2_create_jbd_stats_proc_entry(); 3175 } else { 3176 jbd2_journal_destroy_caches(); 3177 } 3178 return ret; 3179 } 3180 3181 static void __exit journal_exit(void) 3182 { 3183 #ifdef CONFIG_JBD2_DEBUG 3184 int n = atomic_read(&nr_journal_heads); 3185 if (n) 3186 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); 3187 #endif 3188 jbd2_remove_jbd_stats_proc_entry(); 3189 jbd2_journal_destroy_caches(); 3190 } 3191 3192 MODULE_LICENSE("GPL"); 3193 module_init(journal_init); 3194 module_exit(journal_exit); 3195 3196