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_folio); 87 EXPORT_SYMBOL(jbd2_journal_invalidate_folio); 88 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); 89 EXPORT_SYMBOL(jbd2_journal_force_commit); 90 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write); 91 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait); 92 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers); 93 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); 94 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); 95 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); 96 EXPORT_SYMBOL(jbd2_inode_cache); 97 98 static int jbd2_journal_create_slab(size_t slab_size); 99 100 #ifdef CONFIG_JBD2_DEBUG 101 void __jbd2_debug(int level, const char *file, const char *func, 102 unsigned int line, const char *fmt, ...) 103 { 104 struct va_format vaf; 105 va_list args; 106 107 if (level > jbd2_journal_enable_debug) 108 return; 109 va_start(args, fmt); 110 vaf.fmt = fmt; 111 vaf.va = &args; 112 printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf); 113 va_end(args); 114 } 115 #endif 116 117 /* Checksumming functions */ 118 static __be32 jbd2_superblock_csum(journal_superblock_t *sb) 119 { 120 __u32 csum; 121 __be32 old_csum; 122 123 old_csum = sb->s_checksum; 124 sb->s_checksum = 0; 125 csum = jbd2_chksum(~0, (char *)sb, sizeof(journal_superblock_t)); 126 sb->s_checksum = old_csum; 127 128 return cpu_to_be32(csum); 129 } 130 131 /* 132 * Helper function used to manage commit timeouts 133 */ 134 135 static void commit_timeout(struct timer_list *t) 136 { 137 journal_t *journal = timer_container_of(journal, t, j_commit_timer); 138 139 wake_up_process(journal->j_task); 140 } 141 142 /* 143 * kjournald2: The main thread function used to manage a logging device 144 * journal. 145 * 146 * This kernel thread is responsible for two things: 147 * 148 * 1) COMMIT: Every so often we need to commit the current state of the 149 * filesystem to disk. The journal thread is responsible for writing 150 * all of the metadata buffers to disk. If a fast commit is ongoing 151 * journal thread waits until it's done and then continues from 152 * there on. 153 * 154 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all 155 * of the data in that part of the log has been rewritten elsewhere on 156 * the disk. Flushing these old buffers to reclaim space in the log is 157 * known as checkpointing, and this thread is responsible for that job. 158 */ 159 160 static int kjournald2(void *arg) 161 { 162 journal_t *journal = arg; 163 transaction_t *transaction; 164 165 /* 166 * Set up an interval timer which can be used to trigger a commit wakeup 167 * after the commit interval expires 168 */ 169 timer_setup(&journal->j_commit_timer, commit_timeout, 0); 170 171 set_freezable(); 172 173 /* Record that the journal thread is running */ 174 journal->j_task = current; 175 wake_up(&journal->j_wait_done_commit); 176 177 /* 178 * Make sure that no allocations from this kernel thread will ever 179 * recurse to the fs layer because we are responsible for the 180 * transaction commit and any fs involvement might get stuck waiting for 181 * the trasn. commit. 182 */ 183 memalloc_nofs_save(); 184 185 /* 186 * And now, wait forever for commit wakeup events. 187 */ 188 write_lock(&journal->j_state_lock); 189 190 loop: 191 if (journal->j_flags & JBD2_UNMOUNT) 192 goto end_loop; 193 194 jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n", 195 journal->j_commit_sequence, journal->j_commit_request); 196 197 if (journal->j_commit_sequence != journal->j_commit_request) { 198 jbd2_debug(1, "OK, requests differ\n"); 199 write_unlock(&journal->j_state_lock); 200 timer_delete_sync(&journal->j_commit_timer); 201 jbd2_journal_commit_transaction(journal); 202 write_lock(&journal->j_state_lock); 203 goto loop; 204 } 205 206 wake_up(&journal->j_wait_done_commit); 207 if (freezing(current)) { 208 /* 209 * The simpler the better. Flushing journal isn't a 210 * good idea, because that depends on threads that may 211 * be already stopped. 212 */ 213 jbd2_debug(1, "Now suspending kjournald2\n"); 214 write_unlock(&journal->j_state_lock); 215 try_to_freeze(); 216 write_lock(&journal->j_state_lock); 217 } else { 218 /* 219 * We assume on resume that commits are already there, 220 * so we don't sleep 221 */ 222 DEFINE_WAIT(wait); 223 224 prepare_to_wait(&journal->j_wait_commit, &wait, 225 TASK_INTERRUPTIBLE); 226 transaction = journal->j_running_transaction; 227 if (transaction == NULL || 228 time_before(jiffies, transaction->t_expires)) { 229 write_unlock(&journal->j_state_lock); 230 schedule(); 231 write_lock(&journal->j_state_lock); 232 } 233 finish_wait(&journal->j_wait_commit, &wait); 234 } 235 236 jbd2_debug(1, "kjournald2 wakes\n"); 237 238 /* 239 * Were we woken up by a commit wakeup event? 240 */ 241 transaction = journal->j_running_transaction; 242 if (transaction && time_after_eq(jiffies, transaction->t_expires)) { 243 journal->j_commit_request = transaction->t_tid; 244 jbd2_debug(1, "woke because of timeout\n"); 245 } 246 goto loop; 247 248 end_loop: 249 timer_delete_sync(&journal->j_commit_timer); 250 journal->j_task = NULL; 251 wake_up(&journal->j_wait_done_commit); 252 jbd2_debug(1, "Journal thread exiting.\n"); 253 write_unlock(&journal->j_state_lock); 254 return 0; 255 } 256 257 static int jbd2_journal_start_thread(journal_t *journal) 258 { 259 struct task_struct *t; 260 261 t = kthread_run(kjournald2, journal, "jbd2/%s", 262 journal->j_devname); 263 if (IS_ERR(t)) 264 return PTR_ERR(t); 265 266 wait_event(journal->j_wait_done_commit, journal->j_task != NULL); 267 return 0; 268 } 269 270 static void journal_kill_thread(journal_t *journal) 271 { 272 write_lock(&journal->j_state_lock); 273 journal->j_flags |= JBD2_UNMOUNT; 274 275 while (journal->j_task) { 276 write_unlock(&journal->j_state_lock); 277 wake_up(&journal->j_wait_commit); 278 wait_event(journal->j_wait_done_commit, journal->j_task == NULL); 279 write_lock(&journal->j_state_lock); 280 } 281 write_unlock(&journal->j_state_lock); 282 } 283 284 static inline bool jbd2_data_needs_escaping(char *data) 285 { 286 return *((__be32 *)data) == cpu_to_be32(JBD2_MAGIC_NUMBER); 287 } 288 289 static inline void jbd2_data_do_escape(char *data) 290 { 291 *((unsigned int *)data) = 0; 292 } 293 294 /* 295 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. 296 * 297 * Writes a metadata buffer to a given disk block. The actual IO is not 298 * performed but a new buffer_head is constructed which labels the data 299 * to be written with the correct destination disk block. 300 * 301 * Any magic-number escaping which needs to be done will cause a 302 * copy-out here. If the buffer happens to start with the 303 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the 304 * magic number is only written to the log for descripter blocks. In 305 * this case, we copy the data and replace the first word with 0, and we 306 * return a result code which indicates that this buffer needs to be 307 * marked as an escaped buffer in the corresponding log descriptor 308 * block. The missing word can then be restored when the block is read 309 * during recovery. 310 * 311 * If the source buffer has already been modified by a new transaction 312 * since we took the last commit snapshot, we use the frozen copy of 313 * that data for IO. If we end up using the existing buffer_head's data 314 * for the write, then we have to make sure nobody modifies it while the 315 * IO is in progress. do_get_write_access() handles this. 316 * 317 * The function returns a pointer to the buffer_head to be used for IO. 318 * 319 * 320 * Return value: 321 * =0: Finished OK without escape 322 * =1: Finished OK with escape 323 */ 324 325 int jbd2_journal_write_metadata_buffer(transaction_t *transaction, 326 struct journal_head *jh_in, 327 struct buffer_head **bh_out, 328 sector_t blocknr) 329 { 330 int do_escape = 0; 331 struct buffer_head *new_bh; 332 struct folio *new_folio; 333 unsigned int new_offset; 334 struct buffer_head *bh_in = jh2bh(jh_in); 335 journal_t *journal = transaction->t_journal; 336 337 /* 338 * The buffer really shouldn't be locked: only the current committing 339 * transaction is allowed to write it, so nobody else is allowed 340 * to do any IO. 341 * 342 * akpm: except if we're journalling data, and write() output is 343 * also part of a shared mapping, and another thread has 344 * decided to launch a writepage() against this buffer. 345 */ 346 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); 347 348 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); 349 350 /* keep subsequent assertions sane */ 351 atomic_set(&new_bh->b_count, 1); 352 353 spin_lock(&jh_in->b_state_lock); 354 /* 355 * If a new transaction has already done a buffer copy-out, then 356 * we use that version of the data for the commit. 357 */ 358 if (jh_in->b_frozen_data) { 359 new_folio = virt_to_folio(jh_in->b_frozen_data); 360 new_offset = offset_in_folio(new_folio, jh_in->b_frozen_data); 361 do_escape = jbd2_data_needs_escaping(jh_in->b_frozen_data); 362 if (do_escape) 363 jbd2_data_do_escape(jh_in->b_frozen_data); 364 } else { 365 char *tmp; 366 char *mapped_data; 367 368 new_folio = bh_in->b_folio; 369 new_offset = offset_in_folio(new_folio, bh_in->b_data); 370 mapped_data = kmap_local_folio(new_folio, new_offset); 371 /* 372 * Fire data frozen trigger if data already wasn't frozen. Do 373 * this before checking for escaping, as the trigger may modify 374 * the magic offset. If a copy-out happens afterwards, it will 375 * have the correct data in the buffer. 376 */ 377 jbd2_buffer_frozen_trigger(jh_in, mapped_data, 378 jh_in->b_triggers); 379 do_escape = jbd2_data_needs_escaping(mapped_data); 380 kunmap_local(mapped_data); 381 /* 382 * Do we need to do a data copy? 383 */ 384 if (!do_escape) 385 goto escape_done; 386 387 spin_unlock(&jh_in->b_state_lock); 388 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS | __GFP_NOFAIL); 389 spin_lock(&jh_in->b_state_lock); 390 if (jh_in->b_frozen_data) { 391 jbd2_free(tmp, bh_in->b_size); 392 goto copy_done; 393 } 394 395 jh_in->b_frozen_data = tmp; 396 memcpy_from_folio(tmp, new_folio, new_offset, bh_in->b_size); 397 /* 398 * This isn't strictly necessary, as we're using frozen 399 * data for the escaping, but it keeps consistency with 400 * b_frozen_data usage. 401 */ 402 jh_in->b_frozen_triggers = jh_in->b_triggers; 403 404 copy_done: 405 new_folio = virt_to_folio(jh_in->b_frozen_data); 406 new_offset = offset_in_folio(new_folio, jh_in->b_frozen_data); 407 jbd2_data_do_escape(jh_in->b_frozen_data); 408 } 409 410 escape_done: 411 folio_set_bh(new_bh, new_folio, new_offset); 412 new_bh->b_size = bh_in->b_size; 413 new_bh->b_bdev = journal->j_dev; 414 new_bh->b_blocknr = blocknr; 415 new_bh->b_private = bh_in; 416 set_buffer_mapped(new_bh); 417 set_buffer_dirty(new_bh); 418 419 *bh_out = new_bh; 420 421 /* 422 * The to-be-written buffer needs to get moved to the io queue, 423 * and the original buffer whose contents we are shadowing or 424 * copying is moved to the transaction's shadow queue. 425 */ 426 JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); 427 spin_lock(&journal->j_list_lock); 428 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); 429 spin_unlock(&journal->j_list_lock); 430 set_buffer_shadow(bh_in); 431 spin_unlock(&jh_in->b_state_lock); 432 433 return do_escape; 434 } 435 436 /* 437 * Allocation code for the journal file. Manage the space left in the 438 * journal, so that we can begin checkpointing when appropriate. 439 */ 440 441 /* 442 * Called with j_state_lock locked for writing. 443 * Returns true if a transaction commit was started. 444 */ 445 static int __jbd2_log_start_commit(journal_t *journal, tid_t target) 446 { 447 /* Return if the txn has already requested to be committed */ 448 if (journal->j_commit_request == target) 449 return 0; 450 451 /* 452 * The only transaction we can possibly wait upon is the 453 * currently running transaction (if it exists). Otherwise, 454 * the target tid must be an old one. 455 */ 456 if (journal->j_running_transaction && 457 journal->j_running_transaction->t_tid == target) { 458 /* 459 * We want a new commit: OK, mark the request and wakeup the 460 * commit thread. We do _not_ do the commit ourselves. 461 */ 462 463 journal->j_commit_request = target; 464 jbd2_debug(1, "JBD2: requesting commit %u/%u\n", 465 journal->j_commit_request, 466 journal->j_commit_sequence); 467 journal->j_running_transaction->t_requested = jiffies; 468 wake_up(&journal->j_wait_commit); 469 return 1; 470 } else if (!tid_geq(journal->j_commit_request, target)) 471 /* This should never happen, but if it does, preserve 472 the evidence before kjournald goes into a loop and 473 increments j_commit_sequence beyond all recognition. */ 474 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", 475 journal->j_commit_request, 476 journal->j_commit_sequence, 477 target, journal->j_running_transaction ? 478 journal->j_running_transaction->t_tid : 0); 479 return 0; 480 } 481 482 int jbd2_log_start_commit(journal_t *journal, tid_t tid) 483 { 484 int ret; 485 486 write_lock(&journal->j_state_lock); 487 ret = __jbd2_log_start_commit(journal, tid); 488 write_unlock(&journal->j_state_lock); 489 return ret; 490 } 491 492 /* 493 * Force and wait any uncommitted transactions. We can only force the running 494 * transaction if we don't have an active handle, otherwise, we will deadlock. 495 * Returns: <0 in case of error, 496 * 0 if nothing to commit, 497 * 1 if transaction was successfully committed. 498 */ 499 static int __jbd2_journal_force_commit(journal_t *journal) 500 { 501 transaction_t *transaction = NULL; 502 tid_t tid; 503 int need_to_start = 0, ret = 0; 504 505 read_lock(&journal->j_state_lock); 506 if (journal->j_running_transaction && !current->journal_info) { 507 transaction = journal->j_running_transaction; 508 if (!tid_geq(journal->j_commit_request, transaction->t_tid)) 509 need_to_start = 1; 510 } else if (journal->j_committing_transaction) 511 transaction = journal->j_committing_transaction; 512 513 if (!transaction) { 514 /* Nothing to commit */ 515 read_unlock(&journal->j_state_lock); 516 return 0; 517 } 518 tid = transaction->t_tid; 519 read_unlock(&journal->j_state_lock); 520 if (need_to_start) 521 jbd2_log_start_commit(journal, tid); 522 ret = jbd2_log_wait_commit(journal, tid); 523 if (!ret) 524 ret = 1; 525 526 return ret; 527 } 528 529 /** 530 * jbd2_journal_force_commit_nested - Force and wait upon a commit if the 531 * calling process is not within transaction. 532 * 533 * @journal: journal to force 534 * Returns true if progress was made. 535 * 536 * This is used for forcing out undo-protected data which contains 537 * bitmaps, when the fs is running out of space. 538 */ 539 int jbd2_journal_force_commit_nested(journal_t *journal) 540 { 541 int ret; 542 543 ret = __jbd2_journal_force_commit(journal); 544 return ret > 0; 545 } 546 547 /** 548 * jbd2_journal_force_commit() - force any uncommitted transactions 549 * @journal: journal to force 550 * 551 * Caller want unconditional commit. We can only force the running transaction 552 * if we don't have an active handle, otherwise, we will deadlock. 553 */ 554 int jbd2_journal_force_commit(journal_t *journal) 555 { 556 int ret; 557 558 J_ASSERT(!current->journal_info); 559 ret = __jbd2_journal_force_commit(journal); 560 if (ret > 0) 561 ret = 0; 562 return ret; 563 } 564 565 /* 566 * Start a commit of the current running transaction (if any). Returns true 567 * if a transaction is going to be committed (or is currently already 568 * committing), and fills its tid in at *ptid 569 */ 570 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) 571 { 572 int ret = 0; 573 574 write_lock(&journal->j_state_lock); 575 if (journal->j_running_transaction) { 576 tid_t tid = journal->j_running_transaction->t_tid; 577 578 __jbd2_log_start_commit(journal, tid); 579 /* There's a running transaction and we've just made sure 580 * it's commit has been scheduled. */ 581 if (ptid) 582 *ptid = tid; 583 ret = 1; 584 } else if (journal->j_committing_transaction) { 585 /* 586 * If commit has been started, then we have to wait for 587 * completion of that transaction. 588 */ 589 if (ptid) 590 *ptid = journal->j_committing_transaction->t_tid; 591 ret = 1; 592 } 593 write_unlock(&journal->j_state_lock); 594 return ret; 595 } 596 597 /* 598 * Return 1 if a given transaction has not yet sent barrier request 599 * connected with a transaction commit. If 0 is returned, transaction 600 * may or may not have sent the barrier. Used to avoid sending barrier 601 * twice in common cases. 602 */ 603 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) 604 { 605 int ret = 0; 606 transaction_t *commit_trans, *running_trans; 607 608 if (!(journal->j_flags & JBD2_BARRIER)) 609 return 0; 610 read_lock(&journal->j_state_lock); 611 /* Transaction already committed? */ 612 if (tid_geq(journal->j_commit_sequence, tid)) 613 goto out; 614 commit_trans = journal->j_committing_transaction; 615 if (!commit_trans || commit_trans->t_tid != tid) { 616 running_trans = journal->j_running_transaction; 617 /* 618 * The query transaction hasn't started committing, 619 * it must still be running. 620 */ 621 if (WARN_ON_ONCE(!running_trans || 622 running_trans->t_tid != tid)) 623 goto out; 624 625 running_trans->t_need_data_flush = 1; 626 ret = 1; 627 goto out; 628 } 629 /* 630 * Transaction is being committed and we already proceeded to 631 * submitting a flush to fs partition? 632 */ 633 if (journal->j_fs_dev != journal->j_dev) { 634 if (!commit_trans->t_need_data_flush || 635 commit_trans->t_state >= T_COMMIT_DFLUSH) 636 goto out; 637 } else { 638 if (commit_trans->t_state >= T_COMMIT_JFLUSH) 639 goto out; 640 } 641 ret = 1; 642 out: 643 read_unlock(&journal->j_state_lock); 644 return ret; 645 } 646 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); 647 648 /* 649 * Wait for a specified commit to complete. 650 * The caller may not hold the journal lock. 651 */ 652 int jbd2_log_wait_commit(journal_t *journal, tid_t tid) 653 { 654 int err = 0; 655 656 read_lock(&journal->j_state_lock); 657 #ifdef CONFIG_PROVE_LOCKING 658 /* 659 * Some callers make sure transaction is already committing and in that 660 * case we cannot block on open handles anymore. So don't warn in that 661 * case. 662 */ 663 if (tid_gt(tid, journal->j_commit_sequence) && 664 (!journal->j_committing_transaction || 665 journal->j_committing_transaction->t_tid != tid)) { 666 read_unlock(&journal->j_state_lock); 667 jbd2_might_wait_for_commit(journal); 668 read_lock(&journal->j_state_lock); 669 } 670 #endif 671 #ifdef CONFIG_JBD2_DEBUG 672 if (!tid_geq(journal->j_commit_request, tid)) { 673 printk(KERN_ERR 674 "%s: error: j_commit_request=%u, tid=%u\n", 675 __func__, journal->j_commit_request, tid); 676 } 677 #endif 678 while (tid_gt(tid, journal->j_commit_sequence)) { 679 jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n", 680 tid, journal->j_commit_sequence); 681 read_unlock(&journal->j_state_lock); 682 wake_up(&journal->j_wait_commit); 683 wait_event(journal->j_wait_done_commit, 684 !tid_gt(tid, journal->j_commit_sequence)); 685 read_lock(&journal->j_state_lock); 686 } 687 read_unlock(&journal->j_state_lock); 688 689 if (unlikely(is_journal_aborted(journal))) 690 err = -EIO; 691 return err; 692 } 693 694 /* 695 * Start a fast commit. If there's an ongoing fast or full commit wait for 696 * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY 697 * if a fast commit is not needed, either because there's an already a commit 698 * going on or this tid has already been committed. Returns -EINVAL if no jbd2 699 * commit has yet been performed. 700 */ 701 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid) 702 { 703 if (unlikely(is_journal_aborted(journal))) 704 return -EIO; 705 /* 706 * Fast commits only allowed if at least one full commit has 707 * been processed. 708 */ 709 if (!journal->j_stats.ts_tid) 710 return -EINVAL; 711 712 write_lock(&journal->j_state_lock); 713 if (tid_geq(journal->j_commit_sequence, tid)) { 714 write_unlock(&journal->j_state_lock); 715 return -EALREADY; 716 } 717 718 if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING || 719 (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) { 720 DEFINE_WAIT(wait); 721 722 prepare_to_wait(&journal->j_fc_wait, &wait, 723 TASK_UNINTERRUPTIBLE); 724 write_unlock(&journal->j_state_lock); 725 schedule(); 726 finish_wait(&journal->j_fc_wait, &wait); 727 return -EALREADY; 728 } 729 journal->j_flags |= JBD2_FAST_COMMIT_ONGOING; 730 write_unlock(&journal->j_state_lock); 731 732 return 0; 733 } 734 EXPORT_SYMBOL(jbd2_fc_begin_commit); 735 736 /* 737 * Stop a fast commit. If fallback is set, this function starts commit of 738 * TID tid before any other fast commit can start. 739 */ 740 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback) 741 { 742 if (journal->j_fc_cleanup_callback) 743 journal->j_fc_cleanup_callback(journal, 0, tid); 744 write_lock(&journal->j_state_lock); 745 journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING; 746 if (fallback) 747 journal->j_flags |= JBD2_FULL_COMMIT_ONGOING; 748 write_unlock(&journal->j_state_lock); 749 wake_up(&journal->j_fc_wait); 750 if (fallback) 751 return jbd2_complete_transaction(journal, tid); 752 return 0; 753 } 754 755 int jbd2_fc_end_commit(journal_t *journal) 756 { 757 return __jbd2_fc_end_commit(journal, 0, false); 758 } 759 EXPORT_SYMBOL(jbd2_fc_end_commit); 760 761 int jbd2_fc_end_commit_fallback(journal_t *journal) 762 { 763 tid_t tid; 764 765 read_lock(&journal->j_state_lock); 766 tid = journal->j_running_transaction ? 767 journal->j_running_transaction->t_tid : 0; 768 read_unlock(&journal->j_state_lock); 769 return __jbd2_fc_end_commit(journal, tid, true); 770 } 771 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback); 772 773 /* Return 1 when transaction with given tid has already committed. */ 774 int jbd2_transaction_committed(journal_t *journal, tid_t tid) 775 { 776 return tid_geq(READ_ONCE(journal->j_commit_sequence), tid); 777 } 778 EXPORT_SYMBOL(jbd2_transaction_committed); 779 780 /* 781 * When this function returns the transaction corresponding to tid 782 * will be completed. If the transaction has currently running, start 783 * committing that transaction before waiting for it to complete. If 784 * the transaction id is stale, it is by definition already completed, 785 * so just return SUCCESS. 786 */ 787 int jbd2_complete_transaction(journal_t *journal, tid_t tid) 788 { 789 int need_to_wait = 1; 790 791 read_lock(&journal->j_state_lock); 792 if (journal->j_running_transaction && 793 journal->j_running_transaction->t_tid == tid) { 794 if (journal->j_commit_request != tid) { 795 /* transaction not yet started, so request it */ 796 read_unlock(&journal->j_state_lock); 797 jbd2_log_start_commit(journal, tid); 798 goto wait_commit; 799 } 800 } else if (!(journal->j_committing_transaction && 801 journal->j_committing_transaction->t_tid == tid)) 802 need_to_wait = 0; 803 read_unlock(&journal->j_state_lock); 804 if (!need_to_wait) 805 return 0; 806 wait_commit: 807 return jbd2_log_wait_commit(journal, tid); 808 } 809 EXPORT_SYMBOL(jbd2_complete_transaction); 810 811 /* 812 * Log buffer allocation routines: 813 */ 814 815 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) 816 { 817 unsigned long blocknr; 818 819 write_lock(&journal->j_state_lock); 820 J_ASSERT(journal->j_free > 1); 821 822 blocknr = journal->j_head; 823 journal->j_head++; 824 journal->j_free--; 825 if (journal->j_head == journal->j_last) 826 journal->j_head = journal->j_first; 827 write_unlock(&journal->j_state_lock); 828 return jbd2_journal_bmap(journal, blocknr, retp); 829 } 830 831 /* Map one fast commit buffer for use by the file system */ 832 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out) 833 { 834 unsigned long long pblock; 835 unsigned long blocknr; 836 int ret = 0; 837 struct buffer_head *bh; 838 int fc_off; 839 840 *bh_out = NULL; 841 842 if (journal->j_fc_off + journal->j_fc_first >= journal->j_fc_last) 843 return -EINVAL; 844 845 fc_off = journal->j_fc_off; 846 blocknr = journal->j_fc_first + fc_off; 847 journal->j_fc_off++; 848 ret = jbd2_journal_bmap(journal, blocknr, &pblock); 849 if (ret) 850 return ret; 851 852 bh = __getblk(journal->j_dev, pblock, journal->j_blocksize); 853 if (!bh) 854 return -ENOMEM; 855 856 journal->j_fc_wbuf[fc_off] = bh; 857 858 *bh_out = bh; 859 860 return 0; 861 } 862 EXPORT_SYMBOL(jbd2_fc_get_buf); 863 864 /* 865 * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf 866 * for completion. 867 */ 868 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks) 869 { 870 struct buffer_head *bh; 871 int i, j_fc_off; 872 873 j_fc_off = journal->j_fc_off; 874 875 /* 876 * Wait in reverse order to minimize chances of us being woken up before 877 * all IOs have completed 878 */ 879 for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) { 880 bh = journal->j_fc_wbuf[i]; 881 wait_on_buffer(bh); 882 /* 883 * Update j_fc_off so jbd2_fc_release_bufs can release remain 884 * buffer head. 885 */ 886 if (unlikely(!buffer_uptodate(bh))) { 887 journal->j_fc_off = i + 1; 888 return -EIO; 889 } 890 put_bh(bh); 891 journal->j_fc_wbuf[i] = NULL; 892 } 893 894 return 0; 895 } 896 EXPORT_SYMBOL(jbd2_fc_wait_bufs); 897 898 void jbd2_fc_release_bufs(journal_t *journal) 899 { 900 struct buffer_head *bh; 901 int i, j_fc_off; 902 903 j_fc_off = journal->j_fc_off; 904 905 for (i = j_fc_off - 1; i >= 0; i--) { 906 bh = journal->j_fc_wbuf[i]; 907 if (!bh) 908 break; 909 put_bh(bh); 910 journal->j_fc_wbuf[i] = NULL; 911 } 912 } 913 EXPORT_SYMBOL(jbd2_fc_release_bufs); 914 915 /* 916 * Conversion of logical to physical block numbers for the journal 917 * 918 * On external journals the journal blocks are identity-mapped, so 919 * this is a no-op. If needed, we can use j_blk_offset - everything is 920 * ready. 921 */ 922 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, 923 unsigned long long *retp) 924 { 925 int err = 0; 926 unsigned long long ret; 927 sector_t block = blocknr; 928 929 if (journal->j_bmap) { 930 err = journal->j_bmap(journal, &block); 931 if (err == 0) 932 *retp = block; 933 } else if (journal->j_inode) { 934 ret = bmap(journal->j_inode, &block); 935 936 if (ret || !block) { 937 printk(KERN_ALERT "%s: journal block not found " 938 "at offset %lu on %s\n", 939 __func__, blocknr, journal->j_devname); 940 jbd2_journal_abort(journal, ret ? ret : -EFSCORRUPTED); 941 err = -EIO; 942 } else { 943 *retp = block; 944 } 945 946 } else { 947 *retp = blocknr; /* +journal->j_blk_offset */ 948 } 949 return err; 950 } 951 952 /* 953 * We play buffer_head aliasing tricks to write data/metadata blocks to 954 * the journal without copying their contents, but for journal 955 * descriptor blocks we do need to generate bona fide buffers. 956 * 957 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying 958 * the buffer's contents they really should run flush_dcache_folio(bh->b_folio). 959 * But we don't bother doing that, so there will be coherency problems with 960 * mmaps of blockdevs which hold live JBD-controlled filesystems. 961 */ 962 struct buffer_head * 963 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type) 964 { 965 journal_t *journal = transaction->t_journal; 966 struct buffer_head *bh; 967 unsigned long long blocknr; 968 journal_header_t *header; 969 int err; 970 971 err = jbd2_journal_next_log_block(journal, &blocknr); 972 973 if (err) 974 return NULL; 975 976 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); 977 if (!bh) 978 return NULL; 979 atomic_dec(&transaction->t_outstanding_credits); 980 lock_buffer(bh); 981 memset(bh->b_data, 0, journal->j_blocksize); 982 header = (journal_header_t *)bh->b_data; 983 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); 984 header->h_blocktype = cpu_to_be32(type); 985 header->h_sequence = cpu_to_be32(transaction->t_tid); 986 set_buffer_uptodate(bh); 987 unlock_buffer(bh); 988 BUFFER_TRACE(bh, "return this buffer"); 989 return bh; 990 } 991 992 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh) 993 { 994 struct jbd2_journal_block_tail *tail; 995 __u32 csum; 996 997 if (!jbd2_journal_has_csum_v2or3(j)) 998 return; 999 1000 tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - 1001 sizeof(struct jbd2_journal_block_tail)); 1002 tail->t_checksum = 0; 1003 csum = jbd2_chksum(j->j_csum_seed, bh->b_data, j->j_blocksize); 1004 tail->t_checksum = cpu_to_be32(csum); 1005 } 1006 1007 /* 1008 * Return tid of the oldest transaction in the journal and block in the journal 1009 * where the transaction starts. 1010 * 1011 * If the journal is now empty, return which will be the next transaction ID 1012 * we will write and where will that transaction start. 1013 * 1014 * The return value is 0 if journal tail cannot be pushed any further, 1 if 1015 * it can. 1016 */ 1017 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, 1018 unsigned long *block) 1019 { 1020 transaction_t *transaction; 1021 int ret; 1022 1023 read_lock(&journal->j_state_lock); 1024 spin_lock(&journal->j_list_lock); 1025 transaction = journal->j_checkpoint_transactions; 1026 if (transaction) { 1027 *tid = transaction->t_tid; 1028 *block = transaction->t_log_start; 1029 } else if ((transaction = journal->j_committing_transaction) != NULL) { 1030 *tid = transaction->t_tid; 1031 *block = transaction->t_log_start; 1032 } else if ((transaction = journal->j_running_transaction) != NULL) { 1033 *tid = transaction->t_tid; 1034 *block = journal->j_head; 1035 } else { 1036 *tid = journal->j_transaction_sequence; 1037 *block = journal->j_head; 1038 } 1039 ret = tid_gt(*tid, journal->j_tail_sequence); 1040 spin_unlock(&journal->j_list_lock); 1041 read_unlock(&journal->j_state_lock); 1042 1043 return ret; 1044 } 1045 1046 /* 1047 * Update information in journal structure and in on disk journal superblock 1048 * about log tail. This function does not check whether information passed in 1049 * really pushes log tail further. It's responsibility of the caller to make 1050 * sure provided log tail information is valid (e.g. by holding 1051 * j_checkpoint_mutex all the time between computing log tail and calling this 1052 * function as is the case with jbd2_cleanup_journal_tail()). 1053 * 1054 * Requires j_checkpoint_mutex 1055 */ 1056 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1057 { 1058 unsigned long freed; 1059 int ret; 1060 1061 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1062 1063 /* 1064 * We cannot afford for write to remain in drive's caches since as 1065 * soon as we update j_tail, next transaction can start reusing journal 1066 * space and if we lose sb update during power failure we'd replay 1067 * old transaction with possibly newly overwritten data. 1068 */ 1069 ret = jbd2_journal_update_sb_log_tail(journal, tid, block, REQ_FUA); 1070 if (ret) 1071 goto out; 1072 1073 write_lock(&journal->j_state_lock); 1074 freed = block - journal->j_tail; 1075 if (block < journal->j_tail) 1076 freed += journal->j_last - journal->j_first; 1077 1078 trace_jbd2_update_log_tail(journal, tid, block, freed); 1079 jbd2_debug(1, 1080 "Cleaning journal tail from %u to %u (offset %lu), " 1081 "freeing %lu\n", 1082 journal->j_tail_sequence, tid, block, freed); 1083 1084 journal->j_free += freed; 1085 journal->j_tail_sequence = tid; 1086 journal->j_tail = block; 1087 write_unlock(&journal->j_state_lock); 1088 1089 out: 1090 return ret; 1091 } 1092 1093 /* 1094 * This is a variation of __jbd2_update_log_tail which checks for validity of 1095 * provided log tail and locks j_checkpoint_mutex. So it is safe against races 1096 * with other threads updating log tail. 1097 */ 1098 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) 1099 { 1100 mutex_lock_io(&journal->j_checkpoint_mutex); 1101 if (tid_gt(tid, journal->j_tail_sequence)) 1102 __jbd2_update_log_tail(journal, tid, block); 1103 mutex_unlock(&journal->j_checkpoint_mutex); 1104 } 1105 1106 struct jbd2_stats_proc_session { 1107 journal_t *journal; 1108 struct transaction_stats_s *stats; 1109 int start; 1110 int max; 1111 }; 1112 1113 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) 1114 { 1115 return *pos ? NULL : SEQ_START_TOKEN; 1116 } 1117 1118 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) 1119 { 1120 (*pos)++; 1121 return NULL; 1122 } 1123 1124 static int jbd2_seq_info_show(struct seq_file *seq, void *v) 1125 { 1126 struct jbd2_stats_proc_session *s = seq->private; 1127 1128 if (v != SEQ_START_TOKEN) 1129 return 0; 1130 seq_printf(seq, "%lu transactions (%lu requested), " 1131 "each up to %u blocks\n", 1132 s->stats->ts_tid, s->stats->ts_requested, 1133 s->journal->j_max_transaction_buffers); 1134 if (s->stats->ts_tid == 0) 1135 return 0; 1136 seq_printf(seq, "average: \n %ums waiting for transaction\n", 1137 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); 1138 seq_printf(seq, " %ums request delay\n", 1139 (s->stats->ts_requested == 0) ? 0 : 1140 jiffies_to_msecs(s->stats->run.rs_request_delay / 1141 s->stats->ts_requested)); 1142 seq_printf(seq, " %ums running transaction\n", 1143 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); 1144 seq_printf(seq, " %ums transaction was being locked\n", 1145 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); 1146 seq_printf(seq, " %ums flushing data (in ordered mode)\n", 1147 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); 1148 seq_printf(seq, " %ums logging transaction\n", 1149 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); 1150 seq_printf(seq, " %lluus average transaction commit time\n", 1151 div_u64(s->journal->j_average_commit_time, 1000)); 1152 seq_printf(seq, " %lu handles per transaction\n", 1153 s->stats->run.rs_handle_count / s->stats->ts_tid); 1154 seq_printf(seq, " %lu blocks per transaction\n", 1155 s->stats->run.rs_blocks / s->stats->ts_tid); 1156 seq_printf(seq, " %lu logged blocks per transaction\n", 1157 s->stats->run.rs_blocks_logged / s->stats->ts_tid); 1158 return 0; 1159 } 1160 1161 static void jbd2_seq_info_stop(struct seq_file *seq, void *v) 1162 { 1163 } 1164 1165 static const struct seq_operations jbd2_seq_info_ops = { 1166 .start = jbd2_seq_info_start, 1167 .next = jbd2_seq_info_next, 1168 .stop = jbd2_seq_info_stop, 1169 .show = jbd2_seq_info_show, 1170 }; 1171 1172 static int jbd2_seq_info_open(struct inode *inode, struct file *file) 1173 { 1174 journal_t *journal = pde_data(inode); 1175 struct jbd2_stats_proc_session *s; 1176 int rc, size; 1177 1178 s = kmalloc_obj(*s); 1179 if (s == NULL) 1180 return -ENOMEM; 1181 size = sizeof(struct transaction_stats_s); 1182 s->stats = kmalloc(size, GFP_KERNEL); 1183 if (s->stats == NULL) { 1184 kfree(s); 1185 return -ENOMEM; 1186 } 1187 spin_lock(&journal->j_history_lock); 1188 memcpy(s->stats, &journal->j_stats, size); 1189 s->journal = journal; 1190 spin_unlock(&journal->j_history_lock); 1191 1192 rc = seq_open(file, &jbd2_seq_info_ops); 1193 if (rc == 0) { 1194 struct seq_file *m = file->private_data; 1195 m->private = s; 1196 } else { 1197 kfree(s->stats); 1198 kfree(s); 1199 } 1200 return rc; 1201 1202 } 1203 1204 static int jbd2_seq_info_release(struct inode *inode, struct file *file) 1205 { 1206 struct seq_file *seq = file->private_data; 1207 struct jbd2_stats_proc_session *s = seq->private; 1208 kfree(s->stats); 1209 kfree(s); 1210 return seq_release(inode, file); 1211 } 1212 1213 static const struct proc_ops jbd2_info_proc_ops = { 1214 .proc_open = jbd2_seq_info_open, 1215 .proc_read = seq_read, 1216 .proc_lseek = seq_lseek, 1217 .proc_release = jbd2_seq_info_release, 1218 }; 1219 1220 static struct proc_dir_entry *proc_jbd2_stats; 1221 1222 static void jbd2_stats_proc_init(journal_t *journal) 1223 { 1224 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); 1225 if (journal->j_proc_entry) { 1226 proc_create_data("info", S_IRUGO, journal->j_proc_entry, 1227 &jbd2_info_proc_ops, journal); 1228 } 1229 } 1230 1231 static void jbd2_stats_proc_exit(journal_t *journal) 1232 { 1233 remove_proc_entry("info", journal->j_proc_entry); 1234 remove_proc_entry(journal->j_devname, proc_jbd2_stats); 1235 } 1236 1237 /* Minimum size of descriptor tag */ 1238 static int jbd2_min_tag_size(void) 1239 { 1240 /* 1241 * Tag with 32-bit block numbers does not use last four bytes of the 1242 * structure 1243 */ 1244 return sizeof(journal_block_tag_t) - 4; 1245 } 1246 1247 /** 1248 * jbd2_journal_shrink_scan() 1249 * @shrink: shrinker to work on 1250 * @sc: reclaim request to process 1251 * 1252 * Scan the checkpointed buffer on the checkpoint list and release the 1253 * journal_head. 1254 */ 1255 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink, 1256 struct shrink_control *sc) 1257 { 1258 journal_t *journal = shrink->private_data; 1259 unsigned long nr_to_scan = sc->nr_to_scan; 1260 unsigned long nr_shrunk; 1261 unsigned long count; 1262 1263 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1264 trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count); 1265 1266 nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan); 1267 1268 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1269 trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count); 1270 1271 return nr_shrunk; 1272 } 1273 1274 /** 1275 * jbd2_journal_shrink_count() 1276 * @shrink: shrinker to work on 1277 * @sc: reclaim request to process 1278 * 1279 * Count the number of checkpoint buffers on the checkpoint list. 1280 */ 1281 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink, 1282 struct shrink_control *sc) 1283 { 1284 journal_t *journal = shrink->private_data; 1285 unsigned long count; 1286 1287 count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count); 1288 trace_jbd2_shrink_count(journal, sc->nr_to_scan, count); 1289 1290 return count; 1291 } 1292 1293 /* 1294 * If the journal init or create aborts, we need to mark the journal 1295 * superblock as being NULL to prevent the journal destroy from writing 1296 * back a bogus superblock. 1297 */ 1298 static void journal_fail_superblock(journal_t *journal) 1299 { 1300 struct buffer_head *bh = journal->j_sb_buffer; 1301 brelse(bh); 1302 journal->j_sb_buffer = NULL; 1303 } 1304 1305 /* 1306 * Check the superblock for a given journal, performing initial 1307 * validation of the format. 1308 */ 1309 static int journal_check_superblock(journal_t *journal) 1310 { 1311 journal_superblock_t *sb = journal->j_superblock; 1312 int num_fc_blks; 1313 int err = -EINVAL; 1314 1315 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || 1316 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { 1317 printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); 1318 return err; 1319 } 1320 1321 if (be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V1 && 1322 be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V2) { 1323 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); 1324 return err; 1325 } 1326 1327 if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) { 1328 printk(KERN_WARNING "JBD2: journal file too short\n"); 1329 return err; 1330 } 1331 1332 if (be32_to_cpu(sb->s_first) == 0 || 1333 be32_to_cpu(sb->s_first) >= journal->j_total_len) { 1334 printk(KERN_WARNING 1335 "JBD2: Invalid start block of journal: %u\n", 1336 be32_to_cpu(sb->s_first)); 1337 return err; 1338 } 1339 1340 /* 1341 * If this is a V2 superblock, then we have to check the 1342 * features flags on it. 1343 */ 1344 if (!jbd2_format_support_feature(journal)) 1345 return 0; 1346 1347 if ((sb->s_feature_ro_compat & 1348 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || 1349 (sb->s_feature_incompat & 1350 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { 1351 printk(KERN_WARNING "JBD2: Unrecognised features on journal\n"); 1352 return err; 1353 } 1354 1355 num_fc_blks = jbd2_has_feature_fast_commit(journal) ? 1356 jbd2_journal_get_num_fc_blks(sb) : 0; 1357 if (be32_to_cpu(sb->s_maxlen) < JBD2_MIN_JOURNAL_BLOCKS || 1358 be32_to_cpu(sb->s_maxlen) - JBD2_MIN_JOURNAL_BLOCKS < num_fc_blks) { 1359 printk(KERN_ERR "JBD2: journal file too short %u,%d\n", 1360 be32_to_cpu(sb->s_maxlen), num_fc_blks); 1361 return err; 1362 } 1363 1364 if (jbd2_has_feature_csum2(journal) && 1365 jbd2_has_feature_csum3(journal)) { 1366 /* Can't have checksum v2 and v3 at the same time! */ 1367 printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " 1368 "at the same time!\n"); 1369 return err; 1370 } 1371 1372 if (jbd2_journal_has_csum_v2or3(journal) && 1373 jbd2_has_feature_checksum(journal)) { 1374 /* Can't have checksum v1 and v2 on at the same time! */ 1375 printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " 1376 "at the same time!\n"); 1377 return err; 1378 } 1379 1380 if (jbd2_journal_has_csum_v2or3(journal)) { 1381 if (sb->s_checksum_type != JBD2_CRC32C_CHKSUM) { 1382 printk(KERN_ERR "JBD2: Unknown checksum type\n"); 1383 return err; 1384 } 1385 1386 /* Check superblock checksum */ 1387 if (sb->s_checksum != jbd2_superblock_csum(sb)) { 1388 printk(KERN_ERR "JBD2: journal checksum error\n"); 1389 err = -EFSBADCRC; 1390 return err; 1391 } 1392 } 1393 1394 return 0; 1395 } 1396 1397 static int journal_revoke_records_per_block(journal_t *journal) 1398 { 1399 int record_size; 1400 int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t); 1401 1402 if (jbd2_has_feature_64bit(journal)) 1403 record_size = 8; 1404 else 1405 record_size = 4; 1406 1407 if (jbd2_journal_has_csum_v2or3(journal)) 1408 space -= sizeof(struct jbd2_journal_block_tail); 1409 return space / record_size; 1410 } 1411 1412 static int jbd2_journal_get_max_txn_bufs(journal_t *journal) 1413 { 1414 return (journal->j_total_len - journal->j_fc_wbufsize) / 3; 1415 } 1416 1417 /* 1418 * Base amount of descriptor blocks we reserve for each transaction. 1419 */ 1420 static int jbd2_descriptor_blocks_per_trans(journal_t *journal) 1421 { 1422 int tag_space = journal->j_blocksize - sizeof(journal_header_t); 1423 int tags_per_block; 1424 1425 /* Subtract UUID */ 1426 tag_space -= 16; 1427 if (jbd2_journal_has_csum_v2or3(journal)) 1428 tag_space -= sizeof(struct jbd2_journal_block_tail); 1429 /* Commit code leaves a slack space of 16 bytes at the end of block */ 1430 tags_per_block = (tag_space - 16) / journal_tag_bytes(journal); 1431 /* 1432 * Revoke descriptors are accounted separately so we need to reserve 1433 * space for commit block and normal transaction descriptor blocks. 1434 */ 1435 return 1 + DIV_ROUND_UP(jbd2_journal_get_max_txn_bufs(journal), 1436 tags_per_block); 1437 } 1438 1439 /* 1440 * Initialize number of blocks each transaction reserves for its bookkeeping 1441 * and maximum number of blocks a transaction can use. This needs to be called 1442 * after the journal size and the fastcommit area size are initialized. 1443 */ 1444 static void jbd2_journal_init_transaction_limits(journal_t *journal) 1445 { 1446 journal->j_revoke_records_per_block = 1447 journal_revoke_records_per_block(journal); 1448 journal->j_transaction_overhead_buffers = 1449 jbd2_descriptor_blocks_per_trans(journal); 1450 journal->j_max_transaction_buffers = 1451 jbd2_journal_get_max_txn_bufs(journal); 1452 } 1453 1454 /* 1455 * Load the on-disk journal superblock and read the key fields into the 1456 * journal_t. 1457 */ 1458 static int journal_load_superblock(journal_t *journal) 1459 { 1460 int err; 1461 struct buffer_head *bh; 1462 journal_superblock_t *sb; 1463 1464 bh = getblk_unmovable(journal->j_dev, journal->j_blk_offset, 1465 journal->j_blocksize); 1466 if (bh) 1467 err = bh_read(bh, 0); 1468 if (!bh || err < 0) { 1469 pr_err("%s: Cannot read journal superblock\n", __func__); 1470 brelse(bh); 1471 return -EIO; 1472 } 1473 1474 journal->j_sb_buffer = bh; 1475 sb = (journal_superblock_t *)bh->b_data; 1476 journal->j_superblock = sb; 1477 err = journal_check_superblock(journal); 1478 if (err) { 1479 journal_fail_superblock(journal); 1480 return err; 1481 } 1482 1483 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); 1484 journal->j_tail = be32_to_cpu(sb->s_start); 1485 journal->j_first = be32_to_cpu(sb->s_first); 1486 journal->j_errno = be32_to_cpu(sb->s_errno); 1487 journal->j_last = be32_to_cpu(sb->s_maxlen); 1488 1489 if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len) 1490 journal->j_total_len = be32_to_cpu(sb->s_maxlen); 1491 /* Precompute checksum seed for all metadata */ 1492 if (jbd2_journal_has_csum_v2or3(journal)) 1493 journal->j_csum_seed = jbd2_chksum(~0, sb->s_uuid, 1494 sizeof(sb->s_uuid)); 1495 /* After journal features are set, we can compute transaction limits */ 1496 jbd2_journal_init_transaction_limits(journal); 1497 1498 if (jbd2_has_feature_fast_commit(journal)) { 1499 journal->j_fc_last = be32_to_cpu(sb->s_maxlen); 1500 journal->j_last = journal->j_fc_last - 1501 jbd2_journal_get_num_fc_blks(sb); 1502 journal->j_fc_first = journal->j_last + 1; 1503 journal->j_fc_off = 0; 1504 } 1505 1506 return 0; 1507 } 1508 1509 1510 /* 1511 * Management for journal control blocks: functions to create and 1512 * destroy journal_t structures, and to initialise and read existing 1513 * journal blocks from disk. */ 1514 1515 /* The journal_init_common() function creates and fills a journal_t object 1516 * in memory. It calls journal_load_superblock() to load the on-disk journal 1517 * superblock and initialize the journal_t object. 1518 */ 1519 1520 static journal_t *journal_init_common(struct block_device *bdev, 1521 struct block_device *fs_dev, 1522 unsigned long long start, int len, int blocksize) 1523 { 1524 journal_t *journal; 1525 int err; 1526 int n; 1527 1528 journal = kzalloc_obj(*journal); 1529 if (!journal) 1530 return ERR_PTR(-ENOMEM); 1531 1532 lockdep_register_key(&journal->jbd2_trans_commit_key); 1533 journal->j_blocksize = blocksize; 1534 journal->j_dev = bdev; 1535 journal->j_fs_dev = fs_dev; 1536 journal->j_blk_offset = start; 1537 journal->j_total_len = len; 1538 jbd2_init_fs_dev_write_error(journal); 1539 1540 err = journal_load_superblock(journal); 1541 if (err) 1542 goto err_cleanup; 1543 1544 init_waitqueue_head(&journal->j_wait_transaction_locked); 1545 init_waitqueue_head(&journal->j_wait_done_commit); 1546 init_waitqueue_head(&journal->j_wait_commit); 1547 init_waitqueue_head(&journal->j_wait_updates); 1548 init_waitqueue_head(&journal->j_wait_reserved); 1549 init_waitqueue_head(&journal->j_fc_wait); 1550 mutex_init(&journal->j_abort_mutex); 1551 mutex_init(&journal->j_barrier); 1552 mutex_init(&journal->j_checkpoint_mutex); 1553 spin_lock_init(&journal->j_revoke_lock); 1554 spin_lock_init(&journal->j_list_lock); 1555 spin_lock_init(&journal->j_history_lock); 1556 rwlock_init(&journal->j_state_lock); 1557 1558 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); 1559 journal->j_min_batch_time = 0; 1560 journal->j_max_batch_time = 15000; /* 15ms */ 1561 atomic_set(&journal->j_reserved_credits, 0); 1562 lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle", 1563 &journal->jbd2_trans_commit_key, 0); 1564 1565 /* The journal is marked for error until we succeed with recovery! */ 1566 journal->j_flags = JBD2_ABORT; 1567 1568 /* Set up a default-sized revoke table for the new mount. */ 1569 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); 1570 if (err) 1571 goto err_cleanup; 1572 1573 /* 1574 * journal descriptor can store up to n blocks, we need enough 1575 * buffers to write out full descriptor block. 1576 */ 1577 err = -ENOMEM; 1578 n = journal->j_blocksize / jbd2_min_tag_size(); 1579 journal->j_wbufsize = n; 1580 journal->j_fc_wbuf = NULL; 1581 journal->j_wbuf = kmalloc_objs(struct buffer_head *, n); 1582 if (!journal->j_wbuf) 1583 goto err_cleanup; 1584 1585 err = percpu_counter_init(&journal->j_checkpoint_jh_count, 0, 1586 GFP_KERNEL); 1587 if (err) 1588 goto err_cleanup; 1589 1590 journal->j_shrink_transaction = NULL; 1591 1592 journal->j_shrinker = shrinker_alloc(0, "jbd2-journal:(%u:%u)", 1593 MAJOR(bdev->bd_dev), 1594 MINOR(bdev->bd_dev)); 1595 if (!journal->j_shrinker) { 1596 err = -ENOMEM; 1597 goto err_cleanup; 1598 } 1599 1600 journal->j_shrinker->scan_objects = jbd2_journal_shrink_scan; 1601 journal->j_shrinker->count_objects = jbd2_journal_shrink_count; 1602 journal->j_shrinker->private_data = journal; 1603 1604 shrinker_register(journal->j_shrinker); 1605 1606 return journal; 1607 1608 err_cleanup: 1609 percpu_counter_destroy(&journal->j_checkpoint_jh_count); 1610 kfree(journal->j_wbuf); 1611 jbd2_journal_destroy_revoke(journal); 1612 journal_fail_superblock(journal); 1613 lockdep_unregister_key(&journal->jbd2_trans_commit_key); 1614 kfree(journal); 1615 return ERR_PTR(err); 1616 } 1617 1618 /* jbd2_journal_init_dev and jbd2_journal_init_inode: 1619 * 1620 * Create a journal structure assigned some fixed set of disk blocks to 1621 * the journal. We don't actually touch those disk blocks yet, but we 1622 * need to set up all of the mapping information to tell the journaling 1623 * system where the journal blocks are. 1624 * 1625 */ 1626 1627 /** 1628 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure 1629 * @bdev: Block device on which to create the journal 1630 * @fs_dev: Device which hold journalled filesystem for this journal. 1631 * @start: Block nr Start of journal. 1632 * @len: Length of the journal in blocks. 1633 * @blocksize: blocksize of journalling device 1634 * 1635 * Returns: a newly created journal_t * 1636 * 1637 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous 1638 * range of blocks on an arbitrary block device. 1639 * 1640 */ 1641 journal_t *jbd2_journal_init_dev(struct block_device *bdev, 1642 struct block_device *fs_dev, 1643 unsigned long long start, int len, int blocksize) 1644 { 1645 journal_t *journal; 1646 1647 journal = journal_init_common(bdev, fs_dev, start, len, blocksize); 1648 if (IS_ERR(journal)) 1649 return ERR_CAST(journal); 1650 1651 snprintf(journal->j_devname, sizeof(journal->j_devname), 1652 "%pg", journal->j_dev); 1653 strreplace(journal->j_devname, '/', '!'); 1654 jbd2_stats_proc_init(journal); 1655 1656 return journal; 1657 } 1658 1659 /** 1660 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. 1661 * @inode: An inode to create the journal in 1662 * 1663 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as 1664 * the journal. The inode must exist already, must support bmap() and 1665 * must have all data blocks preallocated. 1666 */ 1667 journal_t *jbd2_journal_init_inode(struct inode *inode) 1668 { 1669 journal_t *journal; 1670 sector_t blocknr; 1671 int err = 0; 1672 1673 blocknr = 0; 1674 err = bmap(inode, &blocknr); 1675 if (err || !blocknr) { 1676 pr_err("%s: Cannot locate journal superblock\n", __func__); 1677 return err ? ERR_PTR(err) : ERR_PTR(-EINVAL); 1678 } 1679 1680 jbd2_debug(1, "JBD2: inode %s/%llu, size %lld, bits %d, blksize %ld\n", 1681 inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size, 1682 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); 1683 1684 journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev, 1685 blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits, 1686 inode->i_sb->s_blocksize); 1687 if (IS_ERR(journal)) 1688 return ERR_CAST(journal); 1689 1690 journal->j_inode = inode; 1691 snprintf(journal->j_devname, sizeof(journal->j_devname), 1692 "%pg-%llu", journal->j_dev, journal->j_inode->i_ino); 1693 strreplace(journal->j_devname, '/', '!'); 1694 jbd2_stats_proc_init(journal); 1695 1696 return journal; 1697 } 1698 1699 /* 1700 * Given a journal_t structure, initialise the various fields for 1701 * startup of a new journaling session. We use this both when creating 1702 * a journal, and after recovering an old journal to reset it for 1703 * subsequent use. 1704 */ 1705 1706 static int journal_reset(journal_t *journal) 1707 { 1708 journal_superblock_t *sb = journal->j_superblock; 1709 unsigned long long first, last; 1710 1711 first = be32_to_cpu(sb->s_first); 1712 last = be32_to_cpu(sb->s_maxlen); 1713 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { 1714 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", 1715 first, last); 1716 journal_fail_superblock(journal); 1717 return -EINVAL; 1718 } 1719 1720 journal->j_first = first; 1721 journal->j_last = last; 1722 1723 if (journal->j_head != 0 && journal->j_flags & JBD2_CYCLE_RECORD) { 1724 /* 1725 * Disable the cycled recording mode if the journal head block 1726 * number is not correct. 1727 */ 1728 if (journal->j_head < first || journal->j_head >= last) { 1729 printk(KERN_WARNING "JBD2: Incorrect Journal head block %lu, " 1730 "disable journal_cycle_record\n", 1731 journal->j_head); 1732 journal->j_head = journal->j_first; 1733 } 1734 } else { 1735 journal->j_head = journal->j_first; 1736 } 1737 journal->j_tail = journal->j_head; 1738 journal->j_free = journal->j_last - journal->j_first; 1739 1740 journal->j_tail_sequence = journal->j_transaction_sequence; 1741 journal->j_commit_sequence = journal->j_transaction_sequence - 1; 1742 journal->j_commit_request = journal->j_commit_sequence; 1743 1744 /* 1745 * Now that journal recovery is done, turn fast commits off here. This 1746 * way, if fast commit was enabled before the crash but if now FS has 1747 * disabled it, we don't enable fast commits. 1748 */ 1749 jbd2_clear_feature_fast_commit(journal); 1750 1751 /* 1752 * As a special case, if the on-disk copy is already marked as needing 1753 * no recovery (s_start == 0), then we can safely defer the superblock 1754 * update until the next commit by setting JBD2_FLUSHED. This avoids 1755 * attempting a write to a potential-readonly device. 1756 */ 1757 if (sb->s_start == 0) { 1758 jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb " 1759 "(start %ld, seq %u, errno %d)\n", 1760 journal->j_tail, journal->j_tail_sequence, 1761 journal->j_errno); 1762 journal->j_flags |= JBD2_FLUSHED; 1763 } else { 1764 /* Lock here to make assertions happy... */ 1765 mutex_lock_io(&journal->j_checkpoint_mutex); 1766 /* 1767 * Update log tail information. We use REQ_FUA since new 1768 * transaction will start reusing journal space and so we 1769 * must make sure information about current log tail is on 1770 * disk before that. 1771 */ 1772 jbd2_journal_update_sb_log_tail(journal, 1773 journal->j_tail_sequence, 1774 journal->j_tail, REQ_FUA); 1775 mutex_unlock(&journal->j_checkpoint_mutex); 1776 } 1777 return jbd2_journal_start_thread(journal); 1778 } 1779 1780 /* 1781 * This function expects that the caller will have locked the journal 1782 * buffer head, and will return with it unlocked 1783 */ 1784 static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags) 1785 { 1786 struct buffer_head *bh = journal->j_sb_buffer; 1787 journal_superblock_t *sb = journal->j_superblock; 1788 int ret = 0; 1789 1790 /* Buffer got discarded which means block device got invalidated */ 1791 if (!buffer_mapped(bh)) { 1792 unlock_buffer(bh); 1793 return -EIO; 1794 } 1795 1796 /* 1797 * Always set high priority flags to exempt from block layer's 1798 * QOS policies, e.g. writeback throttle. 1799 */ 1800 write_flags |= JBD2_JOURNAL_REQ_FLAGS; 1801 if (!(journal->j_flags & JBD2_BARRIER)) 1802 write_flags &= ~(REQ_FUA | REQ_PREFLUSH); 1803 1804 trace_jbd2_write_superblock(journal, write_flags); 1805 1806 if (buffer_write_io_error(bh)) { 1807 /* 1808 * Oh, dear. A previous attempt to write the journal 1809 * superblock failed. This could happen because the 1810 * USB device was yanked out. Or it could happen to 1811 * be a transient write error and maybe the block will 1812 * be remapped. Nothing we can do but to retry the 1813 * write and hope for the best. 1814 */ 1815 printk(KERN_ERR "JBD2: previous I/O error detected " 1816 "for journal superblock update for %s.\n", 1817 journal->j_devname); 1818 clear_buffer_write_io_error(bh); 1819 set_buffer_uptodate(bh); 1820 } 1821 if (jbd2_journal_has_csum_v2or3(journal)) 1822 sb->s_checksum = jbd2_superblock_csum(sb); 1823 bh_submit(bh, REQ_OP_WRITE | write_flags, bh_end_write); 1824 wait_on_buffer(bh); 1825 if (buffer_write_io_error(bh)) { 1826 clear_buffer_write_io_error(bh); 1827 set_buffer_uptodate(bh); 1828 ret = -EIO; 1829 } 1830 if (ret) { 1831 printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n", 1832 journal->j_devname); 1833 if (!is_journal_aborted(journal)) 1834 jbd2_journal_abort(journal, ret); 1835 } 1836 1837 return ret; 1838 } 1839 1840 /** 1841 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. 1842 * @journal: The journal to update. 1843 * @tail_tid: TID of the new transaction at the tail of the log 1844 * @tail_block: The first block of the transaction at the tail of the log 1845 * @write_flags: Flags for the journal sb write operation 1846 * 1847 * Update a journal's superblock information about log tail and write it to 1848 * disk, waiting for the IO to complete. 1849 */ 1850 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, 1851 unsigned long tail_block, 1852 blk_opf_t write_flags) 1853 { 1854 journal_superblock_t *sb = journal->j_superblock; 1855 int ret; 1856 1857 if (is_journal_aborted(journal)) 1858 return -EIO; 1859 ret = jbd2_check_fs_dev_write_error(journal); 1860 if (ret) { 1861 jbd2_journal_abort(journal, ret); 1862 return -EIO; 1863 } 1864 1865 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1866 jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", 1867 tail_block, tail_tid); 1868 1869 lock_buffer(journal->j_sb_buffer); 1870 sb->s_sequence = cpu_to_be32(tail_tid); 1871 sb->s_start = cpu_to_be32(tail_block); 1872 1873 ret = jbd2_write_superblock(journal, write_flags); 1874 if (ret) 1875 goto out; 1876 1877 /* Log is no longer empty */ 1878 write_lock(&journal->j_state_lock); 1879 journal->j_flags &= ~JBD2_FLUSHED; 1880 write_unlock(&journal->j_state_lock); 1881 1882 out: 1883 return ret; 1884 } 1885 1886 /** 1887 * jbd2_mark_journal_empty() - Mark on disk journal as empty. 1888 * @journal: The journal to update. 1889 * @write_flags: Flags for the journal sb write operation 1890 * 1891 * Update a journal's dynamic superblock fields to show that journal is empty. 1892 * Write updated superblock to disk waiting for IO to complete. 1893 */ 1894 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags) 1895 { 1896 journal_superblock_t *sb = journal->j_superblock; 1897 bool had_fast_commit = false; 1898 1899 BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); 1900 lock_buffer(journal->j_sb_buffer); 1901 if (sb->s_start == 0) { /* Is it already empty? */ 1902 unlock_buffer(journal->j_sb_buffer); 1903 return; 1904 } 1905 1906 jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n", 1907 journal->j_tail_sequence); 1908 1909 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); 1910 sb->s_start = cpu_to_be32(0); 1911 sb->s_head = cpu_to_be32(journal->j_head); 1912 if (jbd2_has_feature_fast_commit(journal)) { 1913 /* 1914 * When journal is clean, no need to commit fast commit flag and 1915 * make file system incompatible with older kernels. 1916 */ 1917 jbd2_clear_feature_fast_commit(journal); 1918 had_fast_commit = true; 1919 } 1920 1921 jbd2_write_superblock(journal, write_flags); 1922 1923 if (had_fast_commit) 1924 jbd2_set_feature_fast_commit(journal); 1925 1926 /* Log is empty */ 1927 write_lock(&journal->j_state_lock); 1928 journal->j_flags |= JBD2_FLUSHED; 1929 write_unlock(&journal->j_state_lock); 1930 } 1931 1932 /** 1933 * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock) 1934 * @journal: The journal to erase. 1935 * @flags: A discard/zeroout request is sent for each physically contigous 1936 * region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or 1937 * JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation 1938 * to perform. 1939 * 1940 * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes 1941 * will be explicitly written if no hardware offload is available, see 1942 * blkdev_issue_zeroout for more details. 1943 */ 1944 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags) 1945 { 1946 int err = 0; 1947 unsigned long block, log_offset; /* logical */ 1948 unsigned long long phys_block, block_start, block_stop; /* physical */ 1949 loff_t byte_start, byte_stop, byte_count; 1950 1951 /* flags must be set to either discard or zeroout */ 1952 if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags || 1953 ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && 1954 (flags & JBD2_JOURNAL_FLUSH_ZEROOUT))) 1955 return -EINVAL; 1956 1957 if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) && 1958 !bdev_max_discard_sectors(journal->j_dev)) 1959 return -EOPNOTSUPP; 1960 1961 /* 1962 * lookup block mapping and issue discard/zeroout for each 1963 * contiguous region 1964 */ 1965 log_offset = be32_to_cpu(journal->j_superblock->s_first); 1966 block_start = ~0ULL; 1967 for (block = log_offset; block < journal->j_total_len; block++) { 1968 err = jbd2_journal_bmap(journal, block, &phys_block); 1969 if (err) { 1970 pr_err("JBD2: bad block at offset %lu", block); 1971 return err; 1972 } 1973 1974 if (block_start == ~0ULL) 1975 block_stop = block_start = phys_block; 1976 1977 /* 1978 * last block not contiguous with current block, 1979 * process last contiguous region and return to this block on 1980 * next loop 1981 */ 1982 if (phys_block != block_stop) { 1983 block--; 1984 } else { 1985 block_stop++; 1986 /* 1987 * if this isn't the last block of journal, 1988 * no need to process now because next block may also 1989 * be part of this contiguous region 1990 */ 1991 if (block != journal->j_total_len - 1) 1992 continue; 1993 } 1994 1995 /* 1996 * end of contiguous region or this is last block of journal, 1997 * take care of the region 1998 */ 1999 byte_start = block_start * journal->j_blocksize; 2000 byte_stop = block_stop * journal->j_blocksize; 2001 byte_count = (block_stop - block_start) * journal->j_blocksize; 2002 2003 truncate_inode_pages_range(journal->j_dev->bd_mapping, 2004 byte_start, byte_stop - 1); 2005 2006 if (flags & JBD2_JOURNAL_FLUSH_DISCARD) { 2007 err = blkdev_issue_discard(journal->j_dev, 2008 byte_start >> SECTOR_SHIFT, 2009 byte_count >> SECTOR_SHIFT, 2010 GFP_NOFS); 2011 } else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) { 2012 err = blkdev_issue_zeroout(journal->j_dev, 2013 byte_start >> SECTOR_SHIFT, 2014 byte_count >> SECTOR_SHIFT, 2015 GFP_NOFS, 0); 2016 } 2017 2018 if (unlikely(err != 0)) { 2019 pr_err("JBD2: (error %d) unable to wipe journal at physical blocks [%llu, %llu)", 2020 err, block_start, block_stop); 2021 return err; 2022 } 2023 2024 /* reset start and stop after processing a region */ 2025 block_start = ~0ULL; 2026 } 2027 2028 return blkdev_issue_flush(journal->j_dev); 2029 } 2030 2031 /** 2032 * jbd2_journal_update_sb_errno() - Update error in the journal. 2033 * @journal: The journal to update. 2034 * 2035 * Update a journal's errno. Write updated superblock to disk waiting for IO 2036 * to complete. 2037 */ 2038 void jbd2_journal_update_sb_errno(journal_t *journal) 2039 { 2040 journal_superblock_t *sb = journal->j_superblock; 2041 int errcode; 2042 2043 lock_buffer(journal->j_sb_buffer); 2044 errcode = journal->j_errno; 2045 if (errcode == -ESHUTDOWN) 2046 errcode = 0; 2047 jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode); 2048 sb->s_errno = cpu_to_be32(errcode); 2049 2050 jbd2_write_superblock(journal, REQ_FUA); 2051 } 2052 EXPORT_SYMBOL(jbd2_journal_update_sb_errno); 2053 2054 /** 2055 * jbd2_journal_load() - Read journal from disk. 2056 * @journal: Journal to act on. 2057 * 2058 * Given a journal_t structure which tells us which disk blocks contain 2059 * a journal, read the journal from disk to initialise the in-memory 2060 * structures. 2061 */ 2062 int jbd2_journal_load(journal_t *journal) 2063 { 2064 int err; 2065 journal_superblock_t *sb = journal->j_superblock; 2066 2067 /* 2068 * Create a slab for this blocksize 2069 */ 2070 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); 2071 if (err) 2072 return err; 2073 2074 /* Let the recovery code check whether it needs to recover any 2075 * data from the journal. */ 2076 err = jbd2_journal_recover(journal); 2077 if (err) { 2078 pr_warn("JBD2: journal recovery failed\n"); 2079 return err; 2080 } 2081 2082 if (journal->j_failed_commit) { 2083 printk(KERN_ERR "JBD2: journal transaction %u on %s " 2084 "is corrupt.\n", journal->j_failed_commit, 2085 journal->j_devname); 2086 return -EFSCORRUPTED; 2087 } 2088 /* 2089 * clear JBD2_ABORT flag initialized in journal_init_common 2090 * here to update log tail information with the newest seq. 2091 */ 2092 journal->j_flags &= ~JBD2_ABORT; 2093 2094 /* OK, we've finished with the dynamic journal bits: 2095 * reinitialise the dynamic contents of the superblock in memory 2096 * and reset them on disk. */ 2097 err = journal_reset(journal); 2098 if (err) { 2099 pr_warn("JBD2: journal reset failed\n"); 2100 return err; 2101 } 2102 2103 journal->j_flags |= JBD2_LOADED; 2104 return 0; 2105 } 2106 2107 /** 2108 * jbd2_journal_destroy() - Release a journal_t structure. 2109 * @journal: Journal to act on. 2110 * 2111 * Release a journal_t structure once it is no longer in use by the 2112 * journaled object. 2113 * Return <0 if we couldn't clean up the journal. 2114 */ 2115 int jbd2_journal_destroy(journal_t *journal) 2116 { 2117 int err = 0; 2118 2119 /* Wait for the commit thread to wake up and die. */ 2120 journal_kill_thread(journal); 2121 2122 /* Force a final log commit */ 2123 if (journal->j_running_transaction) 2124 jbd2_journal_commit_transaction(journal); 2125 2126 /* Force any old transactions to disk */ 2127 2128 /* Totally anal locking here... */ 2129 spin_lock(&journal->j_list_lock); 2130 while (journal->j_checkpoint_transactions != NULL) { 2131 spin_unlock(&journal->j_list_lock); 2132 mutex_lock_io(&journal->j_checkpoint_mutex); 2133 err = jbd2_log_do_checkpoint(journal); 2134 mutex_unlock(&journal->j_checkpoint_mutex); 2135 /* 2136 * If checkpointing failed, just free the buffers to avoid 2137 * looping forever 2138 */ 2139 if (err) { 2140 jbd2_journal_destroy_checkpoint(journal); 2141 spin_lock(&journal->j_list_lock); 2142 break; 2143 } 2144 spin_lock(&journal->j_list_lock); 2145 } 2146 2147 J_ASSERT(journal->j_running_transaction == NULL); 2148 J_ASSERT(journal->j_committing_transaction == NULL); 2149 J_ASSERT(journal->j_checkpoint_transactions == NULL); 2150 spin_unlock(&journal->j_list_lock); 2151 2152 /* 2153 * OK, all checkpoint transactions have been checked, now check the 2154 * writeback errseq of fs dev and abort the journal if some buffer 2155 * failed to write back to the original location, otherwise the 2156 * filesystem may become inconsistent. 2157 */ 2158 if (!is_journal_aborted(journal)) { 2159 int ret = jbd2_check_fs_dev_write_error(journal); 2160 if (ret) 2161 jbd2_journal_abort(journal, ret); 2162 } 2163 2164 if (journal->j_sb_buffer) { 2165 if (!is_journal_aborted(journal)) { 2166 mutex_lock_io(&journal->j_checkpoint_mutex); 2167 2168 write_lock(&journal->j_state_lock); 2169 journal->j_tail_sequence = 2170 ++journal->j_transaction_sequence; 2171 write_unlock(&journal->j_state_lock); 2172 2173 jbd2_mark_journal_empty(journal, REQ_PREFLUSH | REQ_FUA); 2174 mutex_unlock(&journal->j_checkpoint_mutex); 2175 } else 2176 err = -EIO; 2177 brelse(journal->j_sb_buffer); 2178 } 2179 2180 if (journal->j_shrinker) { 2181 percpu_counter_destroy(&journal->j_checkpoint_jh_count); 2182 shrinker_free(journal->j_shrinker); 2183 } 2184 if (journal->j_proc_entry) 2185 jbd2_stats_proc_exit(journal); 2186 iput(journal->j_inode); 2187 if (journal->j_revoke) 2188 jbd2_journal_destroy_revoke(journal); 2189 kfree(journal->j_fc_wbuf); 2190 kfree(journal->j_wbuf); 2191 lockdep_unregister_key(&journal->jbd2_trans_commit_key); 2192 kfree(journal); 2193 2194 return err; 2195 } 2196 2197 2198 /** 2199 * jbd2_journal_check_used_features() - Check if features specified are used. 2200 * @journal: Journal to check. 2201 * @compat: bitmask of compatible features 2202 * @ro: bitmask of features that force read-only mount 2203 * @incompat: bitmask of incompatible features 2204 * 2205 * Check whether the journal uses all of a given set of 2206 * features. Return true (non-zero) if it does. 2207 **/ 2208 2209 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat, 2210 unsigned long ro, unsigned long incompat) 2211 { 2212 journal_superblock_t *sb; 2213 2214 if (!compat && !ro && !incompat) 2215 return 1; 2216 if (!jbd2_format_support_feature(journal)) 2217 return 0; 2218 2219 sb = journal->j_superblock; 2220 2221 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && 2222 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && 2223 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) 2224 return 1; 2225 2226 return 0; 2227 } 2228 2229 /** 2230 * jbd2_journal_check_available_features() - Check feature set in journalling layer 2231 * @journal: Journal to check. 2232 * @compat: bitmask of compatible features 2233 * @ro: bitmask of features that force read-only mount 2234 * @incompat: bitmask of incompatible features 2235 * 2236 * Check whether the journaling code supports the use of 2237 * all of a given set of features on this journal. Return true 2238 * (non-zero) if it can. */ 2239 2240 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat, 2241 unsigned long ro, unsigned long incompat) 2242 { 2243 if (!compat && !ro && !incompat) 2244 return 1; 2245 2246 if (!jbd2_format_support_feature(journal)) 2247 return 0; 2248 2249 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && 2250 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && 2251 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) 2252 return 1; 2253 2254 return 0; 2255 } 2256 2257 static int 2258 jbd2_journal_initialize_fast_commit(journal_t *journal) 2259 { 2260 journal_superblock_t *sb = journal->j_superblock; 2261 unsigned long long num_fc_blks; 2262 2263 num_fc_blks = jbd2_journal_get_num_fc_blks(sb); 2264 if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS) 2265 return -ENOSPC; 2266 2267 /* Are we called twice? */ 2268 WARN_ON(journal->j_fc_wbuf != NULL); 2269 journal->j_fc_wbuf = kmalloc_objs(struct buffer_head *, num_fc_blks); 2270 if (!journal->j_fc_wbuf) 2271 return -ENOMEM; 2272 2273 journal->j_fc_wbufsize = num_fc_blks; 2274 journal->j_fc_last = journal->j_last; 2275 journal->j_last = journal->j_fc_last - num_fc_blks; 2276 journal->j_fc_first = journal->j_last + 1; 2277 journal->j_fc_off = 0; 2278 journal->j_free = journal->j_last - journal->j_first; 2279 2280 return 0; 2281 } 2282 2283 /** 2284 * jbd2_journal_set_features() - Mark a given journal feature in the superblock 2285 * @journal: Journal to act on. 2286 * @compat: bitmask of compatible features 2287 * @ro: bitmask of features that force read-only mount 2288 * @incompat: bitmask of incompatible features 2289 * 2290 * Mark a given journal feature as present on the 2291 * superblock. Returns true if the requested features could be set. 2292 * 2293 */ 2294 2295 int jbd2_journal_set_features(journal_t *journal, unsigned long compat, 2296 unsigned long ro, unsigned long incompat) 2297 { 2298 #define INCOMPAT_FEATURE_ON(f) \ 2299 ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) 2300 #define COMPAT_FEATURE_ON(f) \ 2301 ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) 2302 journal_superblock_t *sb; 2303 2304 if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) 2305 return 1; 2306 2307 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) 2308 return 0; 2309 2310 /* If enabling v2 checksums, turn on v3 instead */ 2311 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { 2312 incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; 2313 incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; 2314 } 2315 2316 /* Asking for checksumming v3 and v1? Only give them v3. */ 2317 if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && 2318 compat & JBD2_FEATURE_COMPAT_CHECKSUM) 2319 compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; 2320 2321 jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", 2322 compat, ro, incompat); 2323 2324 sb = journal->j_superblock; 2325 2326 if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) { 2327 if (jbd2_journal_initialize_fast_commit(journal)) { 2328 pr_err("JBD2: Cannot enable fast commits.\n"); 2329 return 0; 2330 } 2331 } 2332 2333 lock_buffer(journal->j_sb_buffer); 2334 2335 /* If enabling v3 checksums, update superblock and precompute seed */ 2336 if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { 2337 sb->s_checksum_type = JBD2_CRC32C_CHKSUM; 2338 sb->s_feature_compat &= 2339 ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); 2340 journal->j_csum_seed = jbd2_chksum(~0, sb->s_uuid, 2341 sizeof(sb->s_uuid)); 2342 } 2343 2344 /* If enabling v1 checksums, downgrade superblock */ 2345 if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) 2346 sb->s_feature_incompat &= 2347 ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | 2348 JBD2_FEATURE_INCOMPAT_CSUM_V3); 2349 2350 sb->s_feature_compat |= cpu_to_be32(compat); 2351 sb->s_feature_ro_compat |= cpu_to_be32(ro); 2352 sb->s_feature_incompat |= cpu_to_be32(incompat); 2353 /* 2354 * Update the checksum now so that it is valid even for read-only 2355 * filesystems where jbd2_write_superblock() doesn't get called. 2356 */ 2357 if (jbd2_journal_has_csum_v2or3(journal)) 2358 sb->s_checksum = jbd2_superblock_csum(sb); 2359 unlock_buffer(journal->j_sb_buffer); 2360 jbd2_journal_init_transaction_limits(journal); 2361 2362 return 1; 2363 #undef COMPAT_FEATURE_ON 2364 #undef INCOMPAT_FEATURE_ON 2365 } 2366 2367 /* 2368 * jbd2_journal_clear_features() - Clear a given journal feature in the 2369 * superblock 2370 * @journal: Journal to act on. 2371 * @compat: bitmask of compatible features 2372 * @ro: bitmask of features that force read-only mount 2373 * @incompat: bitmask of incompatible features 2374 * 2375 * Clear a given journal feature as present on the 2376 * superblock. 2377 */ 2378 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, 2379 unsigned long ro, unsigned long incompat) 2380 { 2381 journal_superblock_t *sb; 2382 2383 jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", 2384 compat, ro, incompat); 2385 2386 sb = journal->j_superblock; 2387 2388 lock_buffer(journal->j_sb_buffer); 2389 sb->s_feature_compat &= ~cpu_to_be32(compat); 2390 sb->s_feature_ro_compat &= ~cpu_to_be32(ro); 2391 sb->s_feature_incompat &= ~cpu_to_be32(incompat); 2392 /* 2393 * Update the checksum now so that it is valid even for read-only 2394 * filesystems where jbd2_write_superblock() doesn't get called. 2395 */ 2396 if (jbd2_journal_has_csum_v2or3(journal)) 2397 sb->s_checksum = jbd2_superblock_csum(sb); 2398 unlock_buffer(journal->j_sb_buffer); 2399 jbd2_journal_init_transaction_limits(journal); 2400 } 2401 EXPORT_SYMBOL(jbd2_journal_clear_features); 2402 2403 /** 2404 * jbd2_journal_flush() - Flush journal 2405 * @journal: Journal to act on. 2406 * @flags: optional operation on the journal blocks after the flush (see below) 2407 * 2408 * Flush all data for a given journal to disk and empty the journal. 2409 * Filesystems can use this when remounting readonly to ensure that 2410 * recovery does not need to happen on remount. Optionally, a discard or zeroout 2411 * can be issued on the journal blocks after flushing. 2412 * 2413 * flags: 2414 * JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks 2415 * JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks 2416 */ 2417 int jbd2_journal_flush(journal_t *journal, unsigned int flags) 2418 { 2419 int err = 0; 2420 transaction_t *transaction = NULL; 2421 2422 write_lock(&journal->j_state_lock); 2423 2424 /* Force everything buffered to the log... */ 2425 if (journal->j_running_transaction) { 2426 transaction = journal->j_running_transaction; 2427 __jbd2_log_start_commit(journal, transaction->t_tid); 2428 } else if (journal->j_committing_transaction) 2429 transaction = journal->j_committing_transaction; 2430 2431 /* Wait for the log commit to complete... */ 2432 if (transaction) { 2433 tid_t tid = transaction->t_tid; 2434 2435 write_unlock(&journal->j_state_lock); 2436 jbd2_log_wait_commit(journal, tid); 2437 } else { 2438 write_unlock(&journal->j_state_lock); 2439 } 2440 2441 /* ...and flush everything in the log out to disk. */ 2442 spin_lock(&journal->j_list_lock); 2443 while (!err && journal->j_checkpoint_transactions != NULL) { 2444 spin_unlock(&journal->j_list_lock); 2445 mutex_lock_io(&journal->j_checkpoint_mutex); 2446 err = jbd2_log_do_checkpoint(journal); 2447 mutex_unlock(&journal->j_checkpoint_mutex); 2448 spin_lock(&journal->j_list_lock); 2449 } 2450 spin_unlock(&journal->j_list_lock); 2451 2452 if (is_journal_aborted(journal)) 2453 return -EIO; 2454 2455 mutex_lock_io(&journal->j_checkpoint_mutex); 2456 if (!err) { 2457 err = jbd2_cleanup_journal_tail(journal); 2458 if (err < 0) { 2459 mutex_unlock(&journal->j_checkpoint_mutex); 2460 goto out; 2461 } 2462 err = 0; 2463 } 2464 2465 /* Finally, mark the journal as really needing no recovery. 2466 * This sets s_start==0 in the underlying superblock, which is 2467 * the magic code for a fully-recovered superblock. Any future 2468 * commits of data to the journal will restore the current 2469 * s_start value. */ 2470 jbd2_mark_journal_empty(journal, REQ_FUA); 2471 2472 if (flags) 2473 err = __jbd2_journal_erase(journal, flags); 2474 2475 mutex_unlock(&journal->j_checkpoint_mutex); 2476 write_lock(&journal->j_state_lock); 2477 J_ASSERT(!journal->j_running_transaction); 2478 J_ASSERT(!journal->j_committing_transaction); 2479 J_ASSERT(!journal->j_checkpoint_transactions); 2480 J_ASSERT(journal->j_head == journal->j_tail); 2481 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); 2482 write_unlock(&journal->j_state_lock); 2483 out: 2484 return err; 2485 } 2486 2487 /** 2488 * jbd2_journal_wipe() - Wipe journal contents 2489 * @journal: Journal to act on. 2490 * @write: flag (see below) 2491 * 2492 * Wipe out all of the contents of a journal, safely. This will produce 2493 * a warning if the journal contains any valid recovery information. 2494 * Must be called between journal_init_*() and jbd2_journal_load(). 2495 * 2496 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise 2497 * we merely suppress recovery. 2498 */ 2499 2500 int jbd2_journal_wipe(journal_t *journal, int write) 2501 { 2502 int err; 2503 2504 J_ASSERT (!(journal->j_flags & JBD2_LOADED)); 2505 2506 if (!journal->j_tail) 2507 return 0; 2508 2509 printk(KERN_WARNING "JBD2: %s recovery information on journal\n", 2510 write ? "Clearing" : "Ignoring"); 2511 2512 err = jbd2_journal_skip_recovery(journal); 2513 if (write) { 2514 /* Lock to make assertions happy... */ 2515 mutex_lock_io(&journal->j_checkpoint_mutex); 2516 jbd2_mark_journal_empty(journal, REQ_FUA); 2517 mutex_unlock(&journal->j_checkpoint_mutex); 2518 } 2519 2520 return err; 2521 } 2522 2523 /** 2524 * jbd2_journal_abort () - Shutdown the journal immediately. 2525 * @journal: the journal to shutdown. 2526 * @errno: an error number to record in the journal indicating 2527 * the reason for the shutdown. 2528 * 2529 * Perform a complete, immediate shutdown of the ENTIRE 2530 * journal (not of a single transaction). This operation cannot be 2531 * undone without closing and reopening the journal. 2532 * 2533 * The jbd2_journal_abort function is intended to support higher level error 2534 * recovery mechanisms such as the ext2/ext3 remount-readonly error 2535 * mode. 2536 * 2537 * Journal abort has very specific semantics. Any existing dirty, 2538 * unjournaled buffers in the main filesystem will still be written to 2539 * disk by bdflush, but the journaling mechanism will be suspended 2540 * immediately and no further transaction commits will be honoured. 2541 * 2542 * Any dirty, journaled buffers will be written back to disk without 2543 * hitting the journal. Atomicity cannot be guaranteed on an aborted 2544 * filesystem, but we _do_ attempt to leave as much data as possible 2545 * behind for fsck to use for cleanup. 2546 * 2547 * Any attempt to get a new transaction handle on a journal which is in 2548 * ABORT state will just result in an -EROFS error return. A 2549 * jbd2_journal_stop on an existing handle will return -EIO if we have 2550 * entered abort state during the update. 2551 * 2552 * Recursive transactions are not disturbed by journal abort until the 2553 * final jbd2_journal_stop, which will receive the -EIO error. 2554 * 2555 * Finally, the jbd2_journal_abort call allows the caller to supply an errno 2556 * which will be recorded (if possible) in the journal superblock. This 2557 * allows a client to record failure conditions in the middle of a 2558 * transaction without having to complete the transaction to record the 2559 * failure to disk. ext3_error, for example, now uses this 2560 * functionality. 2561 * 2562 */ 2563 2564 void jbd2_journal_abort(journal_t *journal, int errno) 2565 { 2566 transaction_t *transaction; 2567 2568 /* 2569 * Lock the aborting procedure until everything is done, this avoid 2570 * races between filesystem's error handling flow (e.g. ext4_abort()), 2571 * ensure panic after the error info is written into journal's 2572 * superblock. 2573 */ 2574 mutex_lock(&journal->j_abort_mutex); 2575 /* 2576 * ESHUTDOWN always takes precedence because a file system check 2577 * caused by any other journal abort error is not required after 2578 * a shutdown triggered. 2579 */ 2580 write_lock(&journal->j_state_lock); 2581 if (journal->j_flags & JBD2_ABORT) { 2582 int old_errno = journal->j_errno; 2583 2584 write_unlock(&journal->j_state_lock); 2585 if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) { 2586 journal->j_errno = errno; 2587 jbd2_journal_update_sb_errno(journal); 2588 } 2589 mutex_unlock(&journal->j_abort_mutex); 2590 return; 2591 } 2592 2593 /* 2594 * Mark the abort as occurred and start current running transaction 2595 * to release all journaled buffer. 2596 */ 2597 pr_err("Aborting journal on device %s.\n", journal->j_devname); 2598 2599 journal->j_flags |= JBD2_ABORT; 2600 journal->j_errno = errno; 2601 transaction = journal->j_running_transaction; 2602 if (transaction) 2603 __jbd2_log_start_commit(journal, transaction->t_tid); 2604 write_unlock(&journal->j_state_lock); 2605 2606 /* 2607 * Record errno to the journal super block, so that fsck and jbd2 2608 * layer could realise that a filesystem check is needed. 2609 */ 2610 jbd2_journal_update_sb_errno(journal); 2611 mutex_unlock(&journal->j_abort_mutex); 2612 } 2613 2614 /** 2615 * jbd2_journal_errno() - returns the journal's error state. 2616 * @journal: journal to examine. 2617 * 2618 * This is the errno number set with jbd2_journal_abort(), the last 2619 * time the journal was mounted - if the journal was stopped 2620 * without calling abort this will be 0. 2621 * 2622 * If the journal has been aborted on this mount time -EROFS will 2623 * be returned. 2624 */ 2625 int jbd2_journal_errno(journal_t *journal) 2626 { 2627 int err; 2628 2629 read_lock(&journal->j_state_lock); 2630 if (journal->j_flags & JBD2_ABORT) 2631 err = -EROFS; 2632 else 2633 err = journal->j_errno; 2634 read_unlock(&journal->j_state_lock); 2635 return err; 2636 } 2637 2638 /** 2639 * jbd2_journal_clear_err() - clears the journal's error state 2640 * @journal: journal to act on. 2641 * 2642 * An error must be cleared or acked to take a FS out of readonly 2643 * mode. 2644 */ 2645 int jbd2_journal_clear_err(journal_t *journal) 2646 { 2647 int err = 0; 2648 2649 write_lock(&journal->j_state_lock); 2650 if (journal->j_flags & JBD2_ABORT) 2651 err = -EROFS; 2652 else 2653 journal->j_errno = 0; 2654 write_unlock(&journal->j_state_lock); 2655 return err; 2656 } 2657 2658 /** 2659 * jbd2_journal_ack_err() - Ack journal err. 2660 * @journal: journal to act on. 2661 * 2662 * An error must be cleared or acked to take a FS out of readonly 2663 * mode. 2664 */ 2665 void jbd2_journal_ack_err(journal_t *journal) 2666 { 2667 write_lock(&journal->j_state_lock); 2668 if (journal->j_errno) 2669 journal->j_flags |= JBD2_ACK_ERR; 2670 write_unlock(&journal->j_state_lock); 2671 } 2672 2673 int jbd2_journal_blocks_per_folio(struct inode *inode) 2674 { 2675 return 1 << (PAGE_SHIFT + mapping_max_folio_order(inode->i_mapping) - 2676 inode->i_sb->s_blocksize_bits); 2677 } 2678 2679 /* 2680 * helper functions to deal with 32 or 64bit block numbers. 2681 */ 2682 size_t journal_tag_bytes(journal_t *journal) 2683 { 2684 size_t sz; 2685 2686 if (jbd2_has_feature_csum3(journal)) 2687 return sizeof(journal_block_tag3_t); 2688 2689 sz = sizeof(journal_block_tag_t); 2690 2691 if (jbd2_has_feature_csum2(journal)) 2692 sz += sizeof(__u16); 2693 2694 if (jbd2_has_feature_64bit(journal)) 2695 return sz; 2696 else 2697 return sz - sizeof(__u32); 2698 } 2699 2700 /* 2701 * JBD memory management 2702 * 2703 * These functions are used to allocate block-sized chunks of memory 2704 * used for making copies of buffer_head data. Very often it will be 2705 * page-sized chunks of data, but sometimes it will be in 2706 * sub-page-size chunks. (For example, 16k pages on Power systems 2707 * with a 4k block file system.) For blocks smaller than a page, we 2708 * use a SLAB allocator. There are slab caches for each block size, 2709 * which are allocated at mount time, if necessary, and we only free 2710 * (all of) the slab caches when/if the jbd2 module is unloaded. For 2711 * this reason we don't need to a mutex to protect access to 2712 * jbd2_slab[] allocating or releasing memory; only in 2713 * jbd2_journal_create_slab(). 2714 */ 2715 #define JBD2_MAX_SLABS 8 2716 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; 2717 2718 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { 2719 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", 2720 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" 2721 }; 2722 2723 2724 static void jbd2_journal_destroy_slabs(void) 2725 { 2726 int i; 2727 2728 for (i = 0; i < JBD2_MAX_SLABS; i++) { 2729 kmem_cache_destroy(jbd2_slab[i]); 2730 jbd2_slab[i] = NULL; 2731 } 2732 } 2733 2734 static int jbd2_journal_create_slab(size_t size) 2735 { 2736 static DEFINE_MUTEX(jbd2_slab_create_mutex); 2737 int i = order_base_2(size) - 10; 2738 size_t slab_size; 2739 2740 if (size == PAGE_SIZE) 2741 return 0; 2742 2743 if (i >= JBD2_MAX_SLABS) 2744 return -EINVAL; 2745 2746 if (unlikely(i < 0)) 2747 i = 0; 2748 mutex_lock(&jbd2_slab_create_mutex); 2749 if (jbd2_slab[i]) { 2750 mutex_unlock(&jbd2_slab_create_mutex); 2751 return 0; /* Already created */ 2752 } 2753 2754 slab_size = 1 << (i+10); 2755 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, 2756 slab_size, 0, NULL); 2757 mutex_unlock(&jbd2_slab_create_mutex); 2758 if (!jbd2_slab[i]) { 2759 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); 2760 return -ENOMEM; 2761 } 2762 return 0; 2763 } 2764 2765 static struct kmem_cache *get_slab(size_t size) 2766 { 2767 int i = order_base_2(size) - 10; 2768 2769 BUG_ON(i >= JBD2_MAX_SLABS); 2770 if (unlikely(i < 0)) 2771 i = 0; 2772 BUG_ON(jbd2_slab[i] == NULL); 2773 return jbd2_slab[i]; 2774 } 2775 2776 void *jbd2_alloc(size_t size, gfp_t flags) 2777 { 2778 void *ptr; 2779 2780 BUG_ON(size & (size-1)); /* Must be a power of 2 */ 2781 2782 if (size < PAGE_SIZE) 2783 ptr = kmem_cache_alloc(get_slab(size), flags); 2784 else 2785 ptr = kmalloc(size, flags); 2786 2787 /* Check alignment; SLUB has gotten this wrong in the past, 2788 * and this can lead to user data corruption! */ 2789 BUG_ON(((unsigned long) ptr) & (size-1)); 2790 2791 return ptr; 2792 } 2793 2794 void jbd2_free(void *ptr, size_t size) 2795 { 2796 kfree(ptr); 2797 }; 2798 2799 /* 2800 * Journal_head storage management 2801 */ 2802 static struct kmem_cache *jbd2_journal_head_cache; 2803 #ifdef CONFIG_JBD2_DEBUG 2804 static atomic_t nr_journal_heads = ATOMIC_INIT(0); 2805 #endif 2806 2807 static int __init jbd2_journal_init_journal_head_cache(void) 2808 { 2809 J_ASSERT(!jbd2_journal_head_cache); 2810 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", 2811 sizeof(struct journal_head), 2812 0, /* offset */ 2813 SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU, 2814 NULL); /* ctor */ 2815 if (!jbd2_journal_head_cache) { 2816 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); 2817 return -ENOMEM; 2818 } 2819 return 0; 2820 } 2821 2822 static void jbd2_journal_destroy_journal_head_cache(void) 2823 { 2824 kmem_cache_destroy(jbd2_journal_head_cache); 2825 jbd2_journal_head_cache = NULL; 2826 } 2827 2828 /* 2829 * journal_head splicing and dicing 2830 */ 2831 static struct journal_head *journal_alloc_journal_head(void) 2832 { 2833 struct journal_head *ret; 2834 2835 #ifdef CONFIG_JBD2_DEBUG 2836 atomic_inc(&nr_journal_heads); 2837 #endif 2838 ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); 2839 if (!ret) { 2840 jbd2_debug(1, "out of memory for journal_head\n"); 2841 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); 2842 ret = kmem_cache_zalloc(jbd2_journal_head_cache, 2843 GFP_NOFS | __GFP_NOFAIL); 2844 } 2845 spin_lock_init(&ret->b_state_lock); 2846 return ret; 2847 } 2848 2849 static void journal_free_journal_head(struct journal_head *jh) 2850 { 2851 #ifdef CONFIG_JBD2_DEBUG 2852 atomic_dec(&nr_journal_heads); 2853 memset(jh, JBD2_POISON_FREE, sizeof(*jh)); 2854 #endif 2855 kmem_cache_free(jbd2_journal_head_cache, jh); 2856 } 2857 2858 /* 2859 * A journal_head is attached to a buffer_head whenever JBD has an 2860 * interest in the buffer. 2861 * 2862 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit 2863 * is set. This bit is tested in core kernel code where we need to take 2864 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable 2865 * there. 2866 * 2867 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. 2868 * 2869 * When a buffer has its BH_JBD bit set it is immune from being released by 2870 * core kernel code, mainly via ->b_count. 2871 * 2872 * A journal_head is detached from its buffer_head when the journal_head's 2873 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint 2874 * transaction (b_cp_transaction) hold their references to b_jcount. 2875 * 2876 * Various places in the kernel want to attach a journal_head to a buffer_head 2877 * _before_ attaching the journal_head to a transaction. To protect the 2878 * journal_head in this situation, jbd2_journal_add_journal_head elevates the 2879 * journal_head's b_jcount refcount by one. The caller must call 2880 * jbd2_journal_put_journal_head() to undo this. 2881 * 2882 * So the typical usage would be: 2883 * 2884 * (Attach a journal_head if needed. Increments b_jcount) 2885 * struct journal_head *jh = jbd2_journal_add_journal_head(bh); 2886 * ... 2887 * (Get another reference for transaction) 2888 * jbd2_journal_grab_journal_head(bh); 2889 * jh->b_transaction = xxx; 2890 * (Put original reference) 2891 * jbd2_journal_put_journal_head(jh); 2892 */ 2893 2894 /* 2895 * Give a buffer_head a journal_head. 2896 * 2897 * May sleep. 2898 */ 2899 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) 2900 { 2901 struct journal_head *jh; 2902 struct journal_head *new_jh = NULL; 2903 2904 repeat: 2905 if (!buffer_jbd(bh)) 2906 new_jh = journal_alloc_journal_head(); 2907 2908 jbd_lock_bh_journal_head(bh); 2909 if (buffer_jbd(bh)) { 2910 jh = bh2jh(bh); 2911 } else { 2912 J_ASSERT_BH(bh, 2913 (atomic_read(&bh->b_count) > 0) || 2914 (bh->b_folio && bh->b_folio->mapping)); 2915 2916 if (!new_jh) { 2917 jbd_unlock_bh_journal_head(bh); 2918 goto repeat; 2919 } 2920 2921 jh = new_jh; 2922 new_jh = NULL; /* We consumed it */ 2923 set_buffer_jbd(bh); 2924 bh->b_private = jh; 2925 jh->b_bh = bh; 2926 get_bh(bh); 2927 BUFFER_TRACE(bh, "added journal_head"); 2928 } 2929 jh->b_jcount++; 2930 jbd_unlock_bh_journal_head(bh); 2931 if (new_jh) 2932 journal_free_journal_head(new_jh); 2933 return bh->b_private; 2934 } 2935 2936 /* 2937 * Grab a ref against this buffer_head's journal_head. If it ended up not 2938 * having a journal_head, return NULL 2939 */ 2940 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) 2941 { 2942 struct journal_head *jh = NULL; 2943 2944 jbd_lock_bh_journal_head(bh); 2945 if (buffer_jbd(bh)) { 2946 jh = bh2jh(bh); 2947 jh->b_jcount++; 2948 } 2949 jbd_unlock_bh_journal_head(bh); 2950 return jh; 2951 } 2952 EXPORT_SYMBOL(jbd2_journal_grab_journal_head); 2953 2954 static void __journal_remove_journal_head(struct buffer_head *bh) 2955 { 2956 struct journal_head *jh = bh2jh(bh); 2957 2958 J_ASSERT_JH(jh, jh->b_transaction == NULL); 2959 J_ASSERT_JH(jh, jh->b_next_transaction == NULL); 2960 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); 2961 J_ASSERT_JH(jh, jh->b_jlist == BJ_None); 2962 J_ASSERT_BH(bh, buffer_jbd(bh)); 2963 J_ASSERT_BH(bh, jh2bh(jh) == bh); 2964 BUFFER_TRACE(bh, "remove journal_head"); 2965 2966 /* Unlink before dropping the lock */ 2967 bh->b_private = NULL; 2968 jh->b_bh = NULL; /* debug, really */ 2969 clear_buffer_jbd(bh); 2970 } 2971 2972 static void journal_release_journal_head(struct journal_head *jh, size_t b_size) 2973 { 2974 if (jh->b_frozen_data) { 2975 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); 2976 jbd2_free(jh->b_frozen_data, b_size); 2977 } 2978 if (jh->b_committed_data) { 2979 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); 2980 jbd2_free(jh->b_committed_data, b_size); 2981 } 2982 journal_free_journal_head(jh); 2983 } 2984 2985 /* 2986 * Drop a reference on the passed journal_head. If it fell to zero then 2987 * release the journal_head from the buffer_head. 2988 */ 2989 void jbd2_journal_put_journal_head(struct journal_head *jh) 2990 { 2991 struct buffer_head *bh = jh2bh(jh); 2992 2993 jbd_lock_bh_journal_head(bh); 2994 J_ASSERT_JH(jh, jh->b_jcount > 0); 2995 --jh->b_jcount; 2996 if (!jh->b_jcount) { 2997 __journal_remove_journal_head(bh); 2998 jbd_unlock_bh_journal_head(bh); 2999 journal_release_journal_head(jh, bh->b_size); 3000 __brelse(bh); 3001 } else { 3002 jbd_unlock_bh_journal_head(bh); 3003 } 3004 } 3005 EXPORT_SYMBOL(jbd2_journal_put_journal_head); 3006 3007 /* 3008 * Initialize jbd inode head 3009 */ 3010 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) 3011 { 3012 jinode->i_transaction = NULL; 3013 jinode->i_next_transaction = NULL; 3014 jinode->i_vfs_inode = inode; 3015 jinode->i_flags = 0; 3016 jinode->i_dirty_start_page = 0; 3017 jinode->i_dirty_end_page = 0; 3018 INIT_LIST_HEAD(&jinode->i_list); 3019 } 3020 3021 /* 3022 * Function to be called before we start removing inode from memory (i.e., 3023 * clear_inode() is a fine place to be called from). It removes inode from 3024 * transaction's lists. 3025 */ 3026 void jbd2_journal_release_jbd_inode(journal_t *journal, 3027 struct jbd2_inode *jinode) 3028 { 3029 if (!journal) 3030 return; 3031 restart: 3032 spin_lock(&journal->j_list_lock); 3033 /* Is commit writing out inode - we have to wait */ 3034 if (jinode->i_flags & JI_COMMIT_RUNNING) { 3035 wait_queue_head_t *wq; 3036 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); 3037 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); 3038 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 3039 spin_unlock(&journal->j_list_lock); 3040 schedule(); 3041 finish_wait(wq, &wait.wq_entry); 3042 goto restart; 3043 } 3044 3045 if (jinode->i_transaction) { 3046 list_del(&jinode->i_list); 3047 jinode->i_transaction = NULL; 3048 } 3049 spin_unlock(&journal->j_list_lock); 3050 } 3051 3052 3053 #ifdef CONFIG_PROC_FS 3054 3055 #define JBD2_STATS_PROC_NAME "fs/jbd2" 3056 3057 static void __init jbd2_create_jbd_stats_proc_entry(void) 3058 { 3059 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); 3060 } 3061 3062 static void __exit jbd2_remove_jbd_stats_proc_entry(void) 3063 { 3064 if (proc_jbd2_stats) 3065 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); 3066 } 3067 3068 #else 3069 3070 #define jbd2_create_jbd_stats_proc_entry() do {} while (0) 3071 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) 3072 3073 #endif 3074 3075 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; 3076 3077 static int __init jbd2_journal_init_inode_cache(void) 3078 { 3079 J_ASSERT(!jbd2_inode_cache); 3080 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); 3081 if (!jbd2_inode_cache) { 3082 pr_emerg("JBD2: failed to create inode cache\n"); 3083 return -ENOMEM; 3084 } 3085 return 0; 3086 } 3087 3088 static int __init jbd2_journal_init_handle_cache(void) 3089 { 3090 J_ASSERT(!jbd2_handle_cache); 3091 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); 3092 if (!jbd2_handle_cache) { 3093 printk(KERN_EMERG "JBD2: failed to create handle cache\n"); 3094 return -ENOMEM; 3095 } 3096 return 0; 3097 } 3098 3099 static void jbd2_journal_destroy_inode_cache(void) 3100 { 3101 kmem_cache_destroy(jbd2_inode_cache); 3102 jbd2_inode_cache = NULL; 3103 } 3104 3105 static void jbd2_journal_destroy_handle_cache(void) 3106 { 3107 kmem_cache_destroy(jbd2_handle_cache); 3108 jbd2_handle_cache = NULL; 3109 } 3110 3111 /* 3112 * Module startup and shutdown 3113 */ 3114 3115 static int __init journal_init_caches(void) 3116 { 3117 int ret; 3118 3119 ret = jbd2_journal_init_revoke_record_cache(); 3120 if (ret == 0) 3121 ret = jbd2_journal_init_revoke_table_cache(); 3122 if (ret == 0) 3123 ret = jbd2_journal_init_journal_head_cache(); 3124 if (ret == 0) 3125 ret = jbd2_journal_init_handle_cache(); 3126 if (ret == 0) 3127 ret = jbd2_journal_init_inode_cache(); 3128 if (ret == 0) 3129 ret = jbd2_journal_init_transaction_cache(); 3130 return ret; 3131 } 3132 3133 static void jbd2_journal_destroy_caches(void) 3134 { 3135 jbd2_journal_destroy_revoke_record_cache(); 3136 jbd2_journal_destroy_revoke_table_cache(); 3137 jbd2_journal_destroy_journal_head_cache(); 3138 jbd2_journal_destroy_handle_cache(); 3139 jbd2_journal_destroy_inode_cache(); 3140 jbd2_journal_destroy_transaction_cache(); 3141 jbd2_journal_destroy_slabs(); 3142 } 3143 3144 static int __init journal_init(void) 3145 { 3146 int ret; 3147 3148 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); 3149 3150 ret = journal_init_caches(); 3151 if (ret == 0) { 3152 jbd2_create_jbd_stats_proc_entry(); 3153 } else { 3154 jbd2_journal_destroy_caches(); 3155 } 3156 return ret; 3157 } 3158 3159 static void __exit journal_exit(void) 3160 { 3161 #ifdef CONFIG_JBD2_DEBUG 3162 int n = atomic_read(&nr_journal_heads); 3163 if (n) 3164 printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); 3165 #endif 3166 jbd2_remove_jbd_stats_proc_entry(); 3167 jbd2_journal_destroy_caches(); 3168 } 3169 3170 MODULE_DESCRIPTION("Generic filesystem journal-writing module"); 3171 MODULE_LICENSE("GPL"); 3172 module_init(journal_init); 3173 module_exit(journal_exit); 3174