linux/fs/jbd/journal.c
<<
>>
Prefs
   1/*
   2 * linux/fs/jbd/journal.c
   3 *
   4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
   5 *
   6 * Copyright 1998 Red Hat corp --- All Rights Reserved
   7 *
   8 * This file is part of the Linux kernel and is made available under
   9 * the terms of the GNU General Public License, version 2, or at your
  10 * option, any later version, incorporated herein by reference.
  11 *
  12 * Generic filesystem journal-writing code; part of the ext2fs
  13 * journaling system.
  14 *
  15 * This file manages journals: areas of disk reserved for logging
  16 * transactional updates.  This includes the kernel journaling thread
  17 * which is responsible for scheduling updates to the log.
  18 *
  19 * We do not actually manage the physical storage of the journal in this
  20 * file: that is left to a per-journal policy function, which allows us
  21 * to store the journal within a filesystem-specified area for ext2
  22 * journaling (ext2 can use a reserved inode for storing the log).
  23 */
  24
  25#include <linux/module.h>
  26#include <linux/time.h>
  27#include <linux/fs.h>
  28#include <linux/jbd.h>
  29#include <linux/errno.h>
  30#include <linux/slab.h>
  31#include <linux/init.h>
  32#include <linux/mm.h>
  33#include <linux/freezer.h>
  34#include <linux/pagemap.h>
  35#include <linux/kthread.h>
  36#include <linux/poison.h>
  37#include <linux/proc_fs.h>
  38#include <linux/debugfs.h>
  39#include <linux/ratelimit.h>
  40
  41#define CREATE_TRACE_POINTS
  42#include <trace/events/jbd.h>
  43
  44#include <asm/uaccess.h>
  45#include <asm/page.h>
  46
  47EXPORT_SYMBOL(journal_start);
  48EXPORT_SYMBOL(journal_restart);
  49EXPORT_SYMBOL(journal_extend);
  50EXPORT_SYMBOL(journal_stop);
  51EXPORT_SYMBOL(journal_lock_updates);
  52EXPORT_SYMBOL(journal_unlock_updates);
  53EXPORT_SYMBOL(journal_get_write_access);
  54EXPORT_SYMBOL(journal_get_create_access);
  55EXPORT_SYMBOL(journal_get_undo_access);
  56EXPORT_SYMBOL(journal_dirty_data);
  57EXPORT_SYMBOL(journal_dirty_metadata);
  58EXPORT_SYMBOL(journal_release_buffer);
  59EXPORT_SYMBOL(journal_forget);
  60#if 0
  61EXPORT_SYMBOL(journal_sync_buffer);
  62#endif
  63EXPORT_SYMBOL(journal_flush);
  64EXPORT_SYMBOL(journal_revoke);
  65
  66EXPORT_SYMBOL(journal_init_dev);
  67EXPORT_SYMBOL(journal_init_inode);
  68EXPORT_SYMBOL(journal_update_format);
  69EXPORT_SYMBOL(journal_check_used_features);
  70EXPORT_SYMBOL(journal_check_available_features);
  71EXPORT_SYMBOL(journal_set_features);
  72EXPORT_SYMBOL(journal_create);
  73EXPORT_SYMBOL(journal_load);
  74EXPORT_SYMBOL(journal_destroy);
  75EXPORT_SYMBOL(journal_abort);
  76EXPORT_SYMBOL(journal_errno);
  77EXPORT_SYMBOL(journal_ack_err);
  78EXPORT_SYMBOL(journal_clear_err);
  79EXPORT_SYMBOL(log_wait_commit);
  80EXPORT_SYMBOL(log_start_commit);
  81EXPORT_SYMBOL(journal_start_commit);
  82EXPORT_SYMBOL(journal_force_commit_nested);
  83EXPORT_SYMBOL(journal_wipe);
  84EXPORT_SYMBOL(journal_blocks_per_page);
  85EXPORT_SYMBOL(journal_invalidatepage);
  86EXPORT_SYMBOL(journal_try_to_free_buffers);
  87EXPORT_SYMBOL(journal_force_commit);
  88
  89static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
  90static void __journal_abort_soft (journal_t *journal, int errno);
  91static const char *journal_dev_name(journal_t *journal, char *buffer);
  92
  93/*
  94 * Helper function used to manage commit timeouts
  95 */
  96
  97static void commit_timeout(unsigned long __data)
  98{
  99        struct task_struct * p = (struct task_struct *) __data;
 100
 101        wake_up_process(p);
 102}
 103
 104/*
 105 * kjournald: The main thread function used to manage a logging device
 106 * journal.
 107 *
 108 * This kernel thread is responsible for two things:
 109 *
 110 * 1) COMMIT:  Every so often we need to commit the current state of the
 111 *    filesystem to disk.  The journal thread is responsible for writing
 112 *    all of the metadata buffers to disk.
 113 *
 114 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
 115 *    of the data in that part of the log has been rewritten elsewhere on
 116 *    the disk.  Flushing these old buffers to reclaim space in the log is
 117 *    known as checkpointing, and this thread is responsible for that job.
 118 */
 119
 120static int kjournald(void *arg)
 121{
 122        journal_t *journal = arg;
 123        transaction_t *transaction;
 124
 125        /*
 126         * Set up an interval timer which can be used to trigger a commit wakeup
 127         * after the commit interval expires
 128         */
 129        setup_timer(&journal->j_commit_timer, commit_timeout,
 130                        (unsigned long)current);
 131
 132        set_freezable();
 133
 134        /* Record that the journal thread is running */
 135        journal->j_task = current;
 136        wake_up(&journal->j_wait_done_commit);
 137
 138        printk(KERN_INFO "kjournald starting.  Commit interval %ld seconds\n",
 139                        journal->j_commit_interval / HZ);
 140
 141        /*
 142         * And now, wait forever for commit wakeup events.
 143         */
 144        spin_lock(&journal->j_state_lock);
 145
 146loop:
 147        if (journal->j_flags & JFS_UNMOUNT)
 148                goto end_loop;
 149
 150        jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
 151                journal->j_commit_sequence, journal->j_commit_request);
 152
 153        if (journal->j_commit_sequence != journal->j_commit_request) {
 154                jbd_debug(1, "OK, requests differ\n");
 155                spin_unlock(&journal->j_state_lock);
 156                del_timer_sync(&journal->j_commit_timer);
 157                journal_commit_transaction(journal);
 158                spin_lock(&journal->j_state_lock);
 159                goto loop;
 160        }
 161
 162        wake_up(&journal->j_wait_done_commit);
 163        if (freezing(current)) {
 164                /*
 165                 * The simpler the better. Flushing journal isn't a
 166                 * good idea, because that depends on threads that may
 167                 * be already stopped.
 168                 */
 169                jbd_debug(1, "Now suspending kjournald\n");
 170                spin_unlock(&journal->j_state_lock);
 171                try_to_freeze();
 172                spin_lock(&journal->j_state_lock);
 173        } else {
 174                /*
 175                 * We assume on resume that commits are already there,
 176                 * so we don't sleep
 177                 */
 178                DEFINE_WAIT(wait);
 179                int should_sleep = 1;
 180
 181                prepare_to_wait(&journal->j_wait_commit, &wait,
 182                                TASK_INTERRUPTIBLE);
 183                if (journal->j_commit_sequence != journal->j_commit_request)
 184                        should_sleep = 0;
 185                transaction = journal->j_running_transaction;
 186                if (transaction && time_after_eq(jiffies,
 187                                                transaction->t_expires))
 188                        should_sleep = 0;
 189                if (journal->j_flags & JFS_UNMOUNT)
 190                        should_sleep = 0;
 191                if (should_sleep) {
 192                        spin_unlock(&journal->j_state_lock);
 193                        schedule();
 194                        spin_lock(&journal->j_state_lock);
 195                }
 196                finish_wait(&journal->j_wait_commit, &wait);
 197        }
 198
 199        jbd_debug(1, "kjournald wakes\n");
 200
 201        /*
 202         * Were we woken up by a commit wakeup event?
 203         */
 204        transaction = journal->j_running_transaction;
 205        if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
 206                journal->j_commit_request = transaction->t_tid;
 207                jbd_debug(1, "woke because of timeout\n");
 208        }
 209        goto loop;
 210
 211end_loop:
 212        spin_unlock(&journal->j_state_lock);
 213        del_timer_sync(&journal->j_commit_timer);
 214        journal->j_task = NULL;
 215        wake_up(&journal->j_wait_done_commit);
 216        jbd_debug(1, "Journal thread exiting.\n");
 217        return 0;
 218}
 219
 220static int journal_start_thread(journal_t *journal)
 221{
 222        struct task_struct *t;
 223
 224        t = kthread_run(kjournald, journal, "kjournald");
 225        if (IS_ERR(t))
 226                return PTR_ERR(t);
 227
 228        wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
 229        return 0;
 230}
 231
 232static void journal_kill_thread(journal_t *journal)
 233{
 234        spin_lock(&journal->j_state_lock);
 235        journal->j_flags |= JFS_UNMOUNT;
 236
 237        while (journal->j_task) {
 238                wake_up(&journal->j_wait_commit);
 239                spin_unlock(&journal->j_state_lock);
 240                wait_event(journal->j_wait_done_commit,
 241                                journal->j_task == NULL);
 242                spin_lock(&journal->j_state_lock);
 243        }
 244        spin_unlock(&journal->j_state_lock);
 245}
 246
 247/*
 248 * journal_write_metadata_buffer: write a metadata buffer to the journal.
 249 *
 250 * Writes a metadata buffer to a given disk block.  The actual IO is not
 251 * performed but a new buffer_head is constructed which labels the data
 252 * to be written with the correct destination disk block.
 253 *
 254 * Any magic-number escaping which needs to be done will cause a
 255 * copy-out here.  If the buffer happens to start with the
 256 * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the
 257 * magic number is only written to the log for descripter blocks.  In
 258 * this case, we copy the data and replace the first word with 0, and we
 259 * return a result code which indicates that this buffer needs to be
 260 * marked as an escaped buffer in the corresponding log descriptor
 261 * block.  The missing word can then be restored when the block is read
 262 * during recovery.
 263 *
 264 * If the source buffer has already been modified by a new transaction
 265 * since we took the last commit snapshot, we use the frozen copy of
 266 * that data for IO.  If we end up using the existing buffer_head's data
 267 * for the write, then we *have* to lock the buffer to prevent anyone
 268 * else from using and possibly modifying it while the IO is in
 269 * progress.
 270 *
 271 * The function returns a pointer to the buffer_heads to be used for IO.
 272 *
 273 * We assume that the journal has already been locked in this function.
 274 *
 275 * Return value:
 276 *  <0: Error
 277 * >=0: Finished OK
 278 *
 279 * On success:
 280 * Bit 0 set == escape performed on the data
 281 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
 282 */
 283
 284int journal_write_metadata_buffer(transaction_t *transaction,
 285                                  struct journal_head  *jh_in,
 286                                  struct journal_head **jh_out,
 287                                  unsigned int blocknr)
 288{
 289        int need_copy_out = 0;
 290        int done_copy_out = 0;
 291        int do_escape = 0;
 292        char *mapped_data;
 293        struct buffer_head *new_bh;
 294        struct journal_head *new_jh;
 295        struct page *new_page;
 296        unsigned int new_offset;
 297        struct buffer_head *bh_in = jh2bh(jh_in);
 298        journal_t *journal = transaction->t_journal;
 299
 300        /*
 301         * The buffer really shouldn't be locked: only the current committing
 302         * transaction is allowed to write it, so nobody else is allowed
 303         * to do any IO.
 304         *
 305         * akpm: except if we're journalling data, and write() output is
 306         * also part of a shared mapping, and another thread has
 307         * decided to launch a writepage() against this buffer.
 308         */
 309        J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
 310
 311        new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
 312        /* keep subsequent assertions sane */
 313        new_bh->b_state = 0;
 314        init_buffer(new_bh, NULL, NULL);
 315        atomic_set(&new_bh->b_count, 1);
 316        new_jh = journal_add_journal_head(new_bh);      /* This sleeps */
 317
 318        /*
 319         * If a new transaction has already done a buffer copy-out, then
 320         * we use that version of the data for the commit.
 321         */
 322        jbd_lock_bh_state(bh_in);
 323repeat:
 324        if (jh_in->b_frozen_data) {
 325                done_copy_out = 1;
 326                new_page = virt_to_page(jh_in->b_frozen_data);
 327                new_offset = offset_in_page(jh_in->b_frozen_data);
 328        } else {
 329                new_page = jh2bh(jh_in)->b_page;
 330                new_offset = offset_in_page(jh2bh(jh_in)->b_data);
 331        }
 332
 333        mapped_data = kmap_atomic(new_page);
 334        /*
 335         * Check for escaping
 336         */
 337        if (*((__be32 *)(mapped_data + new_offset)) ==
 338                                cpu_to_be32(JFS_MAGIC_NUMBER)) {
 339                need_copy_out = 1;
 340                do_escape = 1;
 341        }
 342        kunmap_atomic(mapped_data);
 343
 344        /*
 345         * Do we need to do a data copy?
 346         */
 347        if (need_copy_out && !done_copy_out) {
 348                char *tmp;
 349
 350                jbd_unlock_bh_state(bh_in);
 351                tmp = jbd_alloc(bh_in->b_size, GFP_NOFS);
 352                jbd_lock_bh_state(bh_in);
 353                if (jh_in->b_frozen_data) {
 354                        jbd_free(tmp, bh_in->b_size);
 355                        goto repeat;
 356                }
 357
 358                jh_in->b_frozen_data = tmp;
 359                mapped_data = kmap_atomic(new_page);
 360                memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
 361                kunmap_atomic(mapped_data);
 362
 363                new_page = virt_to_page(tmp);
 364                new_offset = offset_in_page(tmp);
 365                done_copy_out = 1;
 366        }
 367
 368        /*
 369         * Did we need to do an escaping?  Now we've done all the
 370         * copying, we can finally do so.
 371         */
 372        if (do_escape) {
 373                mapped_data = kmap_atomic(new_page);
 374                *((unsigned int *)(mapped_data + new_offset)) = 0;
 375                kunmap_atomic(mapped_data);
 376        }
 377
 378        set_bh_page(new_bh, new_page, new_offset);
 379        new_jh->b_transaction = NULL;
 380        new_bh->b_size = jh2bh(jh_in)->b_size;
 381        new_bh->b_bdev = transaction->t_journal->j_dev;
 382        new_bh->b_blocknr = blocknr;
 383        set_buffer_mapped(new_bh);
 384        set_buffer_dirty(new_bh);
 385
 386        *jh_out = new_jh;
 387
 388        /*
 389         * The to-be-written buffer needs to get moved to the io queue,
 390         * and the original buffer whose contents we are shadowing or
 391         * copying is moved to the transaction's shadow queue.
 392         */
 393        JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
 394        spin_lock(&journal->j_list_lock);
 395        __journal_file_buffer(jh_in, transaction, BJ_Shadow);
 396        spin_unlock(&journal->j_list_lock);
 397        jbd_unlock_bh_state(bh_in);
 398
 399        JBUFFER_TRACE(new_jh, "file as BJ_IO");
 400        journal_file_buffer(new_jh, transaction, BJ_IO);
 401
 402        return do_escape | (done_copy_out << 1);
 403}
 404
 405/*
 406 * Allocation code for the journal file.  Manage the space left in the
 407 * journal, so that we can begin checkpointing when appropriate.
 408 */
 409
 410/*
 411 * __log_space_left: Return the number of free blocks left in the journal.
 412 *
 413 * Called with the journal already locked.
 414 *
 415 * Called under j_state_lock
 416 */
 417
 418int __log_space_left(journal_t *journal)
 419{
 420        int left = journal->j_free;
 421
 422        assert_spin_locked(&journal->j_state_lock);
 423
 424        /*
 425         * Be pessimistic here about the number of those free blocks which
 426         * might be required for log descriptor control blocks.
 427         */
 428
 429#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
 430
 431        left -= MIN_LOG_RESERVED_BLOCKS;
 432
 433        if (left <= 0)
 434                return 0;
 435        left -= (left >> 3);
 436        return left;
 437}
 438
 439/*
 440 * Called under j_state_lock.  Returns true if a transaction commit was started.
 441 */
 442int __log_start_commit(journal_t *journal, tid_t target)
 443{
 444        /*
 445         * The only transaction we can possibly wait upon is the
 446         * currently running transaction (if it exists).  Otherwise,
 447         * the target tid must be an old one.
 448         */
 449        if (journal->j_running_transaction &&
 450            journal->j_running_transaction->t_tid == target) {
 451                /*
 452                 * We want a new commit: OK, mark the request and wakeup the
 453                 * commit thread.  We do _not_ do the commit ourselves.
 454                 */
 455
 456                journal->j_commit_request = target;
 457                jbd_debug(1, "JBD: requesting commit %d/%d\n",
 458                          journal->j_commit_request,
 459                          journal->j_commit_sequence);
 460                wake_up(&journal->j_wait_commit);
 461                return 1;
 462        } else if (!tid_geq(journal->j_commit_request, target))
 463                /* This should never happen, but if it does, preserve
 464                   the evidence before kjournald goes into a loop and
 465                   increments j_commit_sequence beyond all recognition. */
 466                WARN_ONCE(1, "jbd: bad log_start_commit: %u %u %u %u\n",
 467                    journal->j_commit_request, journal->j_commit_sequence,
 468                    target, journal->j_running_transaction ?
 469                    journal->j_running_transaction->t_tid : 0);
 470        return 0;
 471}
 472
 473int log_start_commit(journal_t *journal, tid_t tid)
 474{
 475        int ret;
 476
 477        spin_lock(&journal->j_state_lock);
 478        ret = __log_start_commit(journal, tid);
 479        spin_unlock(&journal->j_state_lock);
 480        return ret;
 481}
 482
 483/*
 484 * Force and wait upon a commit if the calling process is not within
 485 * transaction.  This is used for forcing out undo-protected data which contains
 486 * bitmaps, when the fs is running out of space.
 487 *
 488 * We can only force the running transaction if we don't have an active handle;
 489 * otherwise, we will deadlock.
 490 *
 491 * Returns true if a transaction was started.
 492 */
 493int journal_force_commit_nested(journal_t *journal)
 494{
 495        transaction_t *transaction = NULL;
 496        tid_t tid;
 497
 498        spin_lock(&journal->j_state_lock);
 499        if (journal->j_running_transaction && !current->journal_info) {
 500                transaction = journal->j_running_transaction;
 501                __log_start_commit(journal, transaction->t_tid);
 502        } else if (journal->j_committing_transaction)
 503                transaction = journal->j_committing_transaction;
 504
 505        if (!transaction) {
 506                spin_unlock(&journal->j_state_lock);
 507                return 0;       /* Nothing to retry */
 508        }
 509
 510        tid = transaction->t_tid;
 511        spin_unlock(&journal->j_state_lock);
 512        log_wait_commit(journal, tid);
 513        return 1;
 514}
 515
 516/*
 517 * Start a commit of the current running transaction (if any).  Returns true
 518 * if a transaction is going to be committed (or is currently already
 519 * committing), and fills its tid in at *ptid
 520 */
 521int journal_start_commit(journal_t *journal, tid_t *ptid)
 522{
 523        int ret = 0;
 524
 525        spin_lock(&journal->j_state_lock);
 526        if (journal->j_running_transaction) {
 527                tid_t tid = journal->j_running_transaction->t_tid;
 528
 529                __log_start_commit(journal, tid);
 530                /* There's a running transaction and we've just made sure
 531                 * it's commit has been scheduled. */
 532                if (ptid)
 533                        *ptid = tid;
 534                ret = 1;
 535        } else if (journal->j_committing_transaction) {
 536                /*
 537                 * If commit has been started, then we have to wait for
 538                 * completion of that transaction.
 539                 */
 540                if (ptid)
 541                        *ptid = journal->j_committing_transaction->t_tid;
 542                ret = 1;
 543        }
 544        spin_unlock(&journal->j_state_lock);
 545        return ret;
 546}
 547
 548/*
 549 * Wait for a specified commit to complete.
 550 * The caller may not hold the journal lock.
 551 */
 552int log_wait_commit(journal_t *journal, tid_t tid)
 553{
 554        int err = 0;
 555
 556#ifdef CONFIG_JBD_DEBUG
 557        spin_lock(&journal->j_state_lock);
 558        if (!tid_geq(journal->j_commit_request, tid)) {
 559                printk(KERN_EMERG
 560                       "%s: error: j_commit_request=%d, tid=%d\n",
 561                       __func__, journal->j_commit_request, tid);
 562        }
 563        spin_unlock(&journal->j_state_lock);
 564#endif
 565        spin_lock(&journal->j_state_lock);
 566        if (!tid_geq(journal->j_commit_waited, tid))
 567                journal->j_commit_waited = tid;
 568        while (tid_gt(tid, journal->j_commit_sequence)) {
 569                jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
 570                                  tid, journal->j_commit_sequence);
 571                wake_up(&journal->j_wait_commit);
 572                spin_unlock(&journal->j_state_lock);
 573                wait_event(journal->j_wait_done_commit,
 574                                !tid_gt(tid, journal->j_commit_sequence));
 575                spin_lock(&journal->j_state_lock);
 576        }
 577        spin_unlock(&journal->j_state_lock);
 578
 579        if (unlikely(is_journal_aborted(journal))) {
 580                printk(KERN_EMERG "journal commit I/O error\n");
 581                err = -EIO;
 582        }
 583        return err;
 584}
 585
 586/*
 587 * Return 1 if a given transaction has not yet sent barrier request
 588 * connected with a transaction commit. If 0 is returned, transaction
 589 * may or may not have sent the barrier. Used to avoid sending barrier
 590 * twice in common cases.
 591 */
 592int journal_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
 593{
 594        int ret = 0;
 595        transaction_t *commit_trans;
 596
 597        if (!(journal->j_flags & JFS_BARRIER))
 598                return 0;
 599        spin_lock(&journal->j_state_lock);
 600        /* Transaction already committed? */
 601        if (tid_geq(journal->j_commit_sequence, tid))
 602                goto out;
 603        /*
 604         * Transaction is being committed and we already proceeded to
 605         * writing commit record?
 606         */
 607        commit_trans = journal->j_committing_transaction;
 608        if (commit_trans && commit_trans->t_tid == tid &&
 609            commit_trans->t_state >= T_COMMIT_RECORD)
 610                goto out;
 611        ret = 1;
 612out:
 613        spin_unlock(&journal->j_state_lock);
 614        return ret;
 615}
 616EXPORT_SYMBOL(journal_trans_will_send_data_barrier);
 617
 618/*
 619 * Log buffer allocation routines:
 620 */
 621
 622int journal_next_log_block(journal_t *journal, unsigned int *retp)
 623{
 624        unsigned int blocknr;
 625
 626        spin_lock(&journal->j_state_lock);
 627        J_ASSERT(journal->j_free > 1);
 628
 629        blocknr = journal->j_head;
 630        journal->j_head++;
 631        journal->j_free--;
 632        if (journal->j_head == journal->j_last)
 633                journal->j_head = journal->j_first;
 634        spin_unlock(&journal->j_state_lock);
 635        return journal_bmap(journal, blocknr, retp);
 636}
 637
 638/*
 639 * Conversion of logical to physical block numbers for the journal
 640 *
 641 * On external journals the journal blocks are identity-mapped, so
 642 * this is a no-op.  If needed, we can use j_blk_offset - everything is
 643 * ready.
 644 */
 645int journal_bmap(journal_t *journal, unsigned int blocknr,
 646                 unsigned int *retp)
 647{
 648        int err = 0;
 649        unsigned int ret;
 650
 651        if (journal->j_inode) {
 652                ret = bmap(journal->j_inode, blocknr);
 653                if (ret)
 654                        *retp = ret;
 655                else {
 656                        char b[BDEVNAME_SIZE];
 657
 658                        printk(KERN_ALERT "%s: journal block not found "
 659                                        "at offset %u on %s\n",
 660                                __func__,
 661                                blocknr,
 662                                bdevname(journal->j_dev, b));
 663                        err = -EIO;
 664                        __journal_abort_soft(journal, err);
 665                }
 666        } else {
 667                *retp = blocknr; /* +journal->j_blk_offset */
 668        }
 669        return err;
 670}
 671
 672/*
 673 * We play buffer_head aliasing tricks to write data/metadata blocks to
 674 * the journal without copying their contents, but for journal
 675 * descriptor blocks we do need to generate bona fide buffers.
 676 *
 677 * After the caller of journal_get_descriptor_buffer() has finished modifying
 678 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
 679 * But we don't bother doing that, so there will be coherency problems with
 680 * mmaps of blockdevs which hold live JBD-controlled filesystems.
 681 */
 682struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
 683{
 684        struct buffer_head *bh;
 685        unsigned int blocknr;
 686        int err;
 687
 688        err = journal_next_log_block(journal, &blocknr);
 689
 690        if (err)
 691                return NULL;
 692
 693        bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 694        if (!bh)
 695                return NULL;
 696        lock_buffer(bh);
 697        memset(bh->b_data, 0, journal->j_blocksize);
 698        set_buffer_uptodate(bh);
 699        unlock_buffer(bh);
 700        BUFFER_TRACE(bh, "return this buffer");
 701        return journal_add_journal_head(bh);
 702}
 703
 704/*
 705 * Management for journal control blocks: functions to create and
 706 * destroy journal_t structures, and to initialise and read existing
 707 * journal blocks from disk.  */
 708
 709/* First: create and setup a journal_t object in memory.  We initialise
 710 * very few fields yet: that has to wait until we have created the
 711 * journal structures from from scratch, or loaded them from disk. */
 712
 713static journal_t * journal_init_common (void)
 714{
 715        journal_t *journal;
 716        int err;
 717
 718        journal = kzalloc(sizeof(*journal), GFP_KERNEL);
 719        if (!journal)
 720                goto fail;
 721
 722        init_waitqueue_head(&journal->j_wait_transaction_locked);
 723        init_waitqueue_head(&journal->j_wait_logspace);
 724        init_waitqueue_head(&journal->j_wait_done_commit);
 725        init_waitqueue_head(&journal->j_wait_checkpoint);
 726        init_waitqueue_head(&journal->j_wait_commit);
 727        init_waitqueue_head(&journal->j_wait_updates);
 728        mutex_init(&journal->j_checkpoint_mutex);
 729        spin_lock_init(&journal->j_revoke_lock);
 730        spin_lock_init(&journal->j_list_lock);
 731        spin_lock_init(&journal->j_state_lock);
 732
 733        journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE);
 734
 735        /* The journal is marked for error until we succeed with recovery! */
 736        journal->j_flags = JFS_ABORT;
 737
 738        /* Set up a default-sized revoke table for the new mount. */
 739        err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
 740        if (err) {
 741                kfree(journal);
 742                goto fail;
 743        }
 744        return journal;
 745fail:
 746        return NULL;
 747}
 748
 749/* journal_init_dev and journal_init_inode:
 750 *
 751 * Create a journal structure assigned some fixed set of disk blocks to
 752 * the journal.  We don't actually touch those disk blocks yet, but we
 753 * need to set up all of the mapping information to tell the journaling
 754 * system where the journal blocks are.
 755 *
 756 */
 757
 758/**
 759 *  journal_t * journal_init_dev() - creates and initialises a journal structure
 760 *  @bdev: Block device on which to create the journal
 761 *  @fs_dev: Device which hold journalled filesystem for this journal.
 762 *  @start: Block nr Start of journal.
 763 *  @len:  Length of the journal in blocks.
 764 *  @blocksize: blocksize of journalling device
 765 *
 766 *  Returns: a newly created journal_t *
 767 *
 768 *  journal_init_dev creates a journal which maps a fixed contiguous
 769 *  range of blocks on an arbitrary block device.
 770 *
 771 */
 772journal_t * journal_init_dev(struct block_device *bdev,
 773                        struct block_device *fs_dev,
 774                        int start, int len, int blocksize)
 775{
 776        journal_t *journal = journal_init_common();
 777        struct buffer_head *bh;
 778        int n;
 779
 780        if (!journal)
 781                return NULL;
 782
 783        /* journal descriptor can store up to n blocks -bzzz */
 784        journal->j_blocksize = blocksize;
 785        n = journal->j_blocksize / sizeof(journal_block_tag_t);
 786        journal->j_wbufsize = n;
 787        journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
 788        if (!journal->j_wbuf) {
 789                printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
 790                        __func__);
 791                goto out_err;
 792        }
 793        journal->j_dev = bdev;
 794        journal->j_fs_dev = fs_dev;
 795        journal->j_blk_offset = start;
 796        journal->j_maxlen = len;
 797
 798        bh = __getblk(journal->j_dev, start, journal->j_blocksize);
 799        if (!bh) {
 800                printk(KERN_ERR
 801                       "%s: Cannot get buffer for journal superblock\n",
 802                       __func__);
 803                goto out_err;
 804        }
 805        journal->j_sb_buffer = bh;
 806        journal->j_superblock = (journal_superblock_t *)bh->b_data;
 807
 808        return journal;
 809out_err:
 810        kfree(journal->j_wbuf);
 811        kfree(journal);
 812        return NULL;
 813}
 814
 815/**
 816 *  journal_t * journal_init_inode () - creates a journal which maps to a inode.
 817 *  @inode: An inode to create the journal in
 818 *
 819 * journal_init_inode creates a journal which maps an on-disk inode as
 820 * the journal.  The inode must exist already, must support bmap() and
 821 * must have all data blocks preallocated.
 822 */
 823journal_t * journal_init_inode (struct inode *inode)
 824{
 825        struct buffer_head *bh;
 826        journal_t *journal = journal_init_common();
 827        int err;
 828        int n;
 829        unsigned int blocknr;
 830
 831        if (!journal)
 832                return NULL;
 833
 834        journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
 835        journal->j_inode = inode;
 836        jbd_debug(1,
 837                  "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
 838                  journal, inode->i_sb->s_id, inode->i_ino,
 839                  (long long) inode->i_size,
 840                  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
 841
 842        journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
 843        journal->j_blocksize = inode->i_sb->s_blocksize;
 844
 845        /* journal descriptor can store up to n blocks -bzzz */
 846        n = journal->j_blocksize / sizeof(journal_block_tag_t);
 847        journal->j_wbufsize = n;
 848        journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
 849        if (!journal->j_wbuf) {
 850                printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
 851                        __func__);
 852                goto out_err;
 853        }
 854
 855        err = journal_bmap(journal, 0, &blocknr);
 856        /* If that failed, give up */
 857        if (err) {
 858                printk(KERN_ERR "%s: Cannot locate journal superblock\n",
 859                       __func__);
 860                goto out_err;
 861        }
 862
 863        bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 864        if (!bh) {
 865                printk(KERN_ERR
 866                       "%s: Cannot get buffer for journal superblock\n",
 867                       __func__);
 868                goto out_err;
 869        }
 870        journal->j_sb_buffer = bh;
 871        journal->j_superblock = (journal_superblock_t *)bh->b_data;
 872
 873        return journal;
 874out_err:
 875        kfree(journal->j_wbuf);
 876        kfree(journal);
 877        return NULL;
 878}
 879
 880/*
 881 * If the journal init or create aborts, we need to mark the journal
 882 * superblock as being NULL to prevent the journal destroy from writing
 883 * back a bogus superblock.
 884 */
 885static void journal_fail_superblock (journal_t *journal)
 886{
 887        struct buffer_head *bh = journal->j_sb_buffer;
 888        brelse(bh);
 889        journal->j_sb_buffer = NULL;
 890}
 891
 892/*
 893 * Given a journal_t structure, initialise the various fields for
 894 * startup of a new journaling session.  We use this both when creating
 895 * a journal, and after recovering an old journal to reset it for
 896 * subsequent use.
 897 */
 898
 899static int journal_reset(journal_t *journal)
 900{
 901        journal_superblock_t *sb = journal->j_superblock;
 902        unsigned int first, last;
 903
 904        first = be32_to_cpu(sb->s_first);
 905        last = be32_to_cpu(sb->s_maxlen);
 906        if (first + JFS_MIN_JOURNAL_BLOCKS > last + 1) {
 907                printk(KERN_ERR "JBD: Journal too short (blocks %u-%u).\n",
 908                       first, last);
 909                journal_fail_superblock(journal);
 910                return -EINVAL;
 911        }
 912
 913        journal->j_first = first;
 914        journal->j_last = last;
 915
 916        journal->j_head = first;
 917        journal->j_tail = first;
 918        journal->j_free = last - first;
 919
 920        journal->j_tail_sequence = journal->j_transaction_sequence;
 921        journal->j_commit_sequence = journal->j_transaction_sequence - 1;
 922        journal->j_commit_request = journal->j_commit_sequence;
 923
 924        journal->j_max_transaction_buffers = journal->j_maxlen / 4;
 925
 926        /*
 927         * As a special case, if the on-disk copy is already marked as needing
 928         * no recovery (s_start == 0), then we can safely defer the superblock
 929         * update until the next commit by setting JFS_FLUSHED.  This avoids
 930         * attempting a write to a potential-readonly device.
 931         */
 932        if (sb->s_start == 0) {
 933                jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
 934                        "(start %u, seq %d, errno %d)\n",
 935                        journal->j_tail, journal->j_tail_sequence,
 936                        journal->j_errno);
 937                journal->j_flags |= JFS_FLUSHED;
 938        } else {
 939                /* Lock here to make assertions happy... */
 940                mutex_lock(&journal->j_checkpoint_mutex);
 941                /*
 942                 * Update log tail information. We use WRITE_FUA since new
 943                 * transaction will start reusing journal space and so we
 944                 * must make sure information about current log tail is on
 945                 * disk before that.
 946                 */
 947                journal_update_sb_log_tail(journal,
 948                                           journal->j_tail_sequence,
 949                                           journal->j_tail,
 950                                           WRITE_FUA);
 951                mutex_unlock(&journal->j_checkpoint_mutex);
 952        }
 953        return journal_start_thread(journal);
 954}
 955
 956/**
 957 * int journal_create() - Initialise the new journal file
 958 * @journal: Journal to create. This structure must have been initialised
 959 *
 960 * Given a journal_t structure which tells us which disk blocks we can
 961 * use, create a new journal superblock and initialise all of the
 962 * journal fields from scratch.
 963 **/
 964int journal_create(journal_t *journal)
 965{
 966        unsigned int blocknr;
 967        struct buffer_head *bh;
 968        journal_superblock_t *sb;
 969        int i, err;
 970
 971        if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) {
 972                printk (KERN_ERR "Journal length (%d blocks) too short.\n",
 973                        journal->j_maxlen);
 974                journal_fail_superblock(journal);
 975                return -EINVAL;
 976        }
 977
 978        if (journal->j_inode == NULL) {
 979                /*
 980                 * We don't know what block to start at!
 981                 */
 982                printk(KERN_EMERG
 983                       "%s: creation of journal on external device!\n",
 984                       __func__);
 985                BUG();
 986        }
 987
 988        /* Zero out the entire journal on disk.  We cannot afford to
 989           have any blocks on disk beginning with JFS_MAGIC_NUMBER. */
 990        jbd_debug(1, "JBD: Zeroing out journal blocks...\n");
 991        for (i = 0; i < journal->j_maxlen; i++) {
 992                err = journal_bmap(journal, i, &blocknr);
 993                if (err)
 994                        return err;
 995                bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 996                if (unlikely(!bh))
 997                        return -ENOMEM;
 998                lock_buffer(bh);
 999                memset (bh->b_data, 0, journal->j_blocksize);
1000                BUFFER_TRACE(bh, "marking dirty");
1001                mark_buffer_dirty(bh);
1002                BUFFER_TRACE(bh, "marking uptodate");
1003                set_buffer_uptodate(bh);
1004                unlock_buffer(bh);
1005                __brelse(bh);
1006        }
1007
1008        sync_blockdev(journal->j_dev);
1009        jbd_debug(1, "JBD: journal cleared.\n");
1010
1011        /* OK, fill in the initial static fields in the new superblock */
1012        sb = journal->j_superblock;
1013
1014        sb->s_header.h_magic     = cpu_to_be32(JFS_MAGIC_NUMBER);
1015        sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1016
1017        sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
1018        sb->s_maxlen    = cpu_to_be32(journal->j_maxlen);
1019        sb->s_first     = cpu_to_be32(1);
1020
1021        journal->j_transaction_sequence = 1;
1022
1023        journal->j_flags &= ~JFS_ABORT;
1024        journal->j_format_version = 2;
1025
1026        return journal_reset(journal);
1027}
1028
1029static void journal_write_superblock(journal_t *journal, int write_op)
1030{
1031        struct buffer_head *bh = journal->j_sb_buffer;
1032        int ret;
1033
1034        trace_journal_write_superblock(journal, write_op);
1035        if (!(journal->j_flags & JFS_BARRIER))
1036                write_op &= ~(REQ_FUA | REQ_FLUSH);
1037        lock_buffer(bh);
1038        if (buffer_write_io_error(bh)) {
1039                char b[BDEVNAME_SIZE];
1040                /*
1041                 * Oh, dear.  A previous attempt to write the journal
1042                 * superblock failed.  This could happen because the
1043                 * USB device was yanked out.  Or it could happen to
1044                 * be a transient write error and maybe the block will
1045                 * be remapped.  Nothing we can do but to retry the
1046                 * write and hope for the best.
1047                 */
1048                printk(KERN_ERR "JBD: previous I/O error detected "
1049                       "for journal superblock update for %s.\n",
1050                       journal_dev_name(journal, b));
1051                clear_buffer_write_io_error(bh);
1052                set_buffer_uptodate(bh);
1053        }
1054
1055        get_bh(bh);
1056        bh->b_end_io = end_buffer_write_sync;
1057        ret = submit_bh(write_op, bh);
1058        wait_on_buffer(bh);
1059        if (buffer_write_io_error(bh)) {
1060                clear_buffer_write_io_error(bh);
1061                set_buffer_uptodate(bh);
1062                ret = -EIO;
1063        }
1064        if (ret) {
1065                char b[BDEVNAME_SIZE];
1066                printk(KERN_ERR "JBD: Error %d detected "
1067                       "when updating journal superblock for %s.\n",
1068                       ret, journal_dev_name(journal, b));
1069        }
1070}
1071
1072/**
1073 * journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1074 * @journal: The journal to update.
1075 * @tail_tid: TID of the new transaction at the tail of the log
1076 * @tail_block: The first block of the transaction at the tail of the log
1077 * @write_op: With which operation should we write the journal sb
1078 *
1079 * Update a journal's superblock information about log tail and write it to
1080 * disk, waiting for the IO to complete.
1081 */
1082void journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1083                                unsigned int tail_block, int write_op)
1084{
1085        journal_superblock_t *sb = journal->j_superblock;
1086
1087        BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1088        jbd_debug(1,"JBD: updating superblock (start %u, seq %u)\n",
1089                  tail_block, tail_tid);
1090
1091        sb->s_sequence = cpu_to_be32(tail_tid);
1092        sb->s_start    = cpu_to_be32(tail_block);
1093
1094        journal_write_superblock(journal, write_op);
1095
1096        /* Log is no longer empty */
1097        spin_lock(&journal->j_state_lock);
1098        WARN_ON(!sb->s_sequence);
1099        journal->j_flags &= ~JFS_FLUSHED;
1100        spin_unlock(&journal->j_state_lock);
1101}
1102
1103/**
1104 * mark_journal_empty() - Mark on disk journal as empty.
1105 * @journal: The journal to update.
1106 *
1107 * Update a journal's dynamic superblock fields to show that journal is empty.
1108 * Write updated superblock to disk waiting for IO to complete.
1109 */
1110static void mark_journal_empty(journal_t *journal)
1111{
1112        journal_superblock_t *sb = journal->j_superblock;
1113
1114        BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1115        spin_lock(&journal->j_state_lock);
1116        /* Is it already empty? */
1117        if (sb->s_start == 0) {
1118                spin_unlock(&journal->j_state_lock);
1119                return;
1120        }
1121        jbd_debug(1, "JBD: Marking journal as empty (seq %d)\n",
1122                  journal->j_tail_sequence);
1123
1124        sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1125        sb->s_start    = cpu_to_be32(0);
1126        spin_unlock(&journal->j_state_lock);
1127
1128        journal_write_superblock(journal, WRITE_FUA);
1129
1130        spin_lock(&journal->j_state_lock);
1131        /* Log is empty */
1132        journal->j_flags |= JFS_FLUSHED;
1133        spin_unlock(&journal->j_state_lock);
1134}
1135
1136/**
1137 * journal_update_sb_errno() - Update error in the journal.
1138 * @journal: The journal to update.
1139 *
1140 * Update a journal's errno.  Write updated superblock to disk waiting for IO
1141 * to complete.
1142 */
1143static void journal_update_sb_errno(journal_t *journal)
1144{
1145        journal_superblock_t *sb = journal->j_superblock;
1146
1147        spin_lock(&journal->j_state_lock);
1148        jbd_debug(1, "JBD: updating superblock error (errno %d)\n",
1149                  journal->j_errno);
1150        sb->s_errno = cpu_to_be32(journal->j_errno);
1151        spin_unlock(&journal->j_state_lock);
1152
1153        journal_write_superblock(journal, WRITE_SYNC);
1154}
1155
1156/*
1157 * Read the superblock for a given journal, performing initial
1158 * validation of the format.
1159 */
1160
1161static int journal_get_superblock(journal_t *journal)
1162{
1163        struct buffer_head *bh;
1164        journal_superblock_t *sb;
1165        int err = -EIO;
1166
1167        bh = journal->j_sb_buffer;
1168
1169        J_ASSERT(bh != NULL);
1170        if (!buffer_uptodate(bh)) {
1171                ll_rw_block(READ, 1, &bh);
1172                wait_on_buffer(bh);
1173                if (!buffer_uptodate(bh)) {
1174                        printk (KERN_ERR
1175                                "JBD: IO error reading journal superblock\n");
1176                        goto out;
1177                }
1178        }
1179
1180        sb = journal->j_superblock;
1181
1182        err = -EINVAL;
1183
1184        if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
1185            sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1186                printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1187                goto out;
1188        }
1189
1190        switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1191        case JFS_SUPERBLOCK_V1:
1192                journal->j_format_version = 1;
1193                break;
1194        case JFS_SUPERBLOCK_V2:
1195                journal->j_format_version = 2;
1196                break;
1197        default:
1198                printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1199                goto out;
1200        }
1201
1202        if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1203                journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1204        else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1205                printk (KERN_WARNING "JBD: journal file too short\n");
1206                goto out;
1207        }
1208
1209        if (be32_to_cpu(sb->s_first) == 0 ||
1210            be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1211                printk(KERN_WARNING
1212                        "JBD: Invalid start block of journal: %u\n",
1213                        be32_to_cpu(sb->s_first));
1214                goto out;
1215        }
1216
1217        return 0;
1218
1219out:
1220        journal_fail_superblock(journal);
1221        return err;
1222}
1223
1224/*
1225 * Load the on-disk journal superblock and read the key fields into the
1226 * journal_t.
1227 */
1228
1229static int load_superblock(journal_t *journal)
1230{
1231        int err;
1232        journal_superblock_t *sb;
1233
1234        err = journal_get_superblock(journal);
1235        if (err)
1236                return err;
1237
1238        sb = journal->j_superblock;
1239
1240        journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1241        journal->j_tail = be32_to_cpu(sb->s_start);
1242        journal->j_first = be32_to_cpu(sb->s_first);
1243        journal->j_last = be32_to_cpu(sb->s_maxlen);
1244        journal->j_errno = be32_to_cpu(sb->s_errno);
1245
1246        return 0;
1247}
1248
1249
1250/**
1251 * int journal_load() - Read journal from disk.
1252 * @journal: Journal to act on.
1253 *
1254 * Given a journal_t structure which tells us which disk blocks contain
1255 * a journal, read the journal from disk to initialise the in-memory
1256 * structures.
1257 */
1258int journal_load(journal_t *journal)
1259{
1260        int err;
1261        journal_superblock_t *sb;
1262
1263        err = load_superblock(journal);
1264        if (err)
1265                return err;
1266
1267        sb = journal->j_superblock;
1268        /* If this is a V2 superblock, then we have to check the
1269         * features flags on it. */
1270
1271        if (journal->j_format_version >= 2) {
1272                if ((sb->s_feature_ro_compat &
1273                     ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) ||
1274                    (sb->s_feature_incompat &
1275                     ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) {
1276                        printk (KERN_WARNING
1277                                "JBD: Unrecognised features on journal\n");
1278                        return -EINVAL;
1279                }
1280        }
1281
1282        /* Let the recovery code check whether it needs to recover any
1283         * data from the journal. */
1284        if (journal_recover(journal))
1285                goto recovery_error;
1286
1287        /* OK, we've finished with the dynamic journal bits:
1288         * reinitialise the dynamic contents of the superblock in memory
1289         * and reset them on disk. */
1290        if (journal_reset(journal))
1291                goto recovery_error;
1292
1293        journal->j_flags &= ~JFS_ABORT;
1294        journal->j_flags |= JFS_LOADED;
1295        return 0;
1296
1297recovery_error:
1298        printk (KERN_WARNING "JBD: recovery failed\n");
1299        return -EIO;
1300}
1301
1302/**
1303 * void journal_destroy() - Release a journal_t structure.
1304 * @journal: Journal to act on.
1305 *
1306 * Release a journal_t structure once it is no longer in use by the
1307 * journaled object.
1308 * Return <0 if we couldn't clean up the journal.
1309 */
1310int journal_destroy(journal_t *journal)
1311{
1312        int err = 0;
1313
1314        
1315        /* Wait for the commit thread to wake up and die. */
1316        journal_kill_thread(journal);
1317
1318        /* Force a final log commit */
1319        if (journal->j_running_transaction)
1320                journal_commit_transaction(journal);
1321
1322        /* Force any old transactions to disk */
1323
1324        /* We cannot race with anybody but must keep assertions happy */
1325        mutex_lock(&journal->j_checkpoint_mutex);
1326        /* Totally anal locking here... */
1327        spin_lock(&journal->j_list_lock);
1328        while (journal->j_checkpoint_transactions != NULL) {
1329                spin_unlock(&journal->j_list_lock);
1330                log_do_checkpoint(journal);
1331                spin_lock(&journal->j_list_lock);
1332        }
1333
1334        J_ASSERT(journal->j_running_transaction == NULL);
1335        J_ASSERT(journal->j_committing_transaction == NULL);
1336        J_ASSERT(journal->j_checkpoint_transactions == NULL);
1337        spin_unlock(&journal->j_list_lock);
1338
1339        if (journal->j_sb_buffer) {
1340                if (!is_journal_aborted(journal)) {
1341                        journal->j_tail_sequence =
1342                                ++journal->j_transaction_sequence;
1343                        mark_journal_empty(journal);
1344                } else
1345                        err = -EIO;
1346                brelse(journal->j_sb_buffer);
1347        }
1348        mutex_unlock(&journal->j_checkpoint_mutex);
1349
1350        if (journal->j_inode)
1351                iput(journal->j_inode);
1352        if (journal->j_revoke)
1353                journal_destroy_revoke(journal);
1354        kfree(journal->j_wbuf);
1355        kfree(journal);
1356
1357        return err;
1358}
1359
1360
1361/**
1362 *int journal_check_used_features () - Check if features specified are used.
1363 * @journal: Journal to check.
1364 * @compat: bitmask of compatible features
1365 * @ro: bitmask of features that force read-only mount
1366 * @incompat: bitmask of incompatible features
1367 *
1368 * Check whether the journal uses all of a given set of
1369 * features.  Return true (non-zero) if it does.
1370 **/
1371
1372int journal_check_used_features (journal_t *journal, unsigned long compat,
1373                                 unsigned long ro, unsigned long incompat)
1374{
1375        journal_superblock_t *sb;
1376
1377        if (!compat && !ro && !incompat)
1378                return 1;
1379        if (journal->j_format_version == 1)
1380                return 0;
1381
1382        sb = journal->j_superblock;
1383
1384        if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1385            ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1386            ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1387                return 1;
1388
1389        return 0;
1390}
1391
1392/**
1393 * int journal_check_available_features() - Check feature set in journalling layer
1394 * @journal: Journal to check.
1395 * @compat: bitmask of compatible features
1396 * @ro: bitmask of features that force read-only mount
1397 * @incompat: bitmask of incompatible features
1398 *
1399 * Check whether the journaling code supports the use of
1400 * all of a given set of features on this journal.  Return true
1401 * (non-zero) if it can. */
1402
1403int journal_check_available_features (journal_t *journal, unsigned long compat,
1404                                      unsigned long ro, unsigned long incompat)
1405{
1406        if (!compat && !ro && !incompat)
1407                return 1;
1408
1409        /* We can support any known requested features iff the
1410         * superblock is in version 2.  Otherwise we fail to support any
1411         * extended sb features. */
1412
1413        if (journal->j_format_version != 2)
1414                return 0;
1415
1416        if ((compat   & JFS_KNOWN_COMPAT_FEATURES) == compat &&
1417            (ro       & JFS_KNOWN_ROCOMPAT_FEATURES) == ro &&
1418            (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat)
1419                return 1;
1420
1421        return 0;
1422}
1423
1424/**
1425 * int journal_set_features () - Mark a given journal feature in the superblock
1426 * @journal: Journal to act on.
1427 * @compat: bitmask of compatible features
1428 * @ro: bitmask of features that force read-only mount
1429 * @incompat: bitmask of incompatible features
1430 *
1431 * Mark a given journal feature as present on the
1432 * superblock.  Returns true if the requested features could be set.
1433 *
1434 */
1435
1436int journal_set_features (journal_t *journal, unsigned long compat,
1437                          unsigned long ro, unsigned long incompat)
1438{
1439        journal_superblock_t *sb;
1440
1441        if (journal_check_used_features(journal, compat, ro, incompat))
1442                return 1;
1443
1444        if (!journal_check_available_features(journal, compat, ro, incompat))
1445                return 0;
1446
1447        jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1448                  compat, ro, incompat);
1449
1450        sb = journal->j_superblock;
1451
1452        sb->s_feature_compat    |= cpu_to_be32(compat);
1453        sb->s_feature_ro_compat |= cpu_to_be32(ro);
1454        sb->s_feature_incompat  |= cpu_to_be32(incompat);
1455
1456        return 1;
1457}
1458
1459
1460/**
1461 * int journal_update_format () - Update on-disk journal structure.
1462 * @journal: Journal to act on.
1463 *
1464 * Given an initialised but unloaded journal struct, poke about in the
1465 * on-disk structure to update it to the most recent supported version.
1466 */
1467int journal_update_format (journal_t *journal)
1468{
1469        journal_superblock_t *sb;
1470        int err;
1471
1472        err = journal_get_superblock(journal);
1473        if (err)
1474                return err;
1475
1476        sb = journal->j_superblock;
1477
1478        switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1479        case JFS_SUPERBLOCK_V2:
1480                return 0;
1481        case JFS_SUPERBLOCK_V1:
1482                return journal_convert_superblock_v1(journal, sb);
1483        default:
1484                break;
1485        }
1486        return -EINVAL;
1487}
1488
1489static int journal_convert_superblock_v1(journal_t *journal,
1490                                         journal_superblock_t *sb)
1491{
1492        int offset, blocksize;
1493        struct buffer_head *bh;
1494
1495        printk(KERN_WARNING
1496                "JBD: Converting superblock from version 1 to 2.\n");
1497
1498        /* Pre-initialise new fields to zero */
1499        offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1500        blocksize = be32_to_cpu(sb->s_blocksize);
1501        memset(&sb->s_feature_compat, 0, blocksize-offset);
1502
1503        sb->s_nr_users = cpu_to_be32(1);
1504        sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
1505        journal->j_format_version = 2;
1506
1507        bh = journal->j_sb_buffer;
1508        BUFFER_TRACE(bh, "marking dirty");
1509        mark_buffer_dirty(bh);
1510        sync_dirty_buffer(bh);
1511        return 0;
1512}
1513
1514
1515/**
1516 * int journal_flush () - Flush journal
1517 * @journal: Journal to act on.
1518 *
1519 * Flush all data for a given journal to disk and empty the journal.
1520 * Filesystems can use this when remounting readonly to ensure that
1521 * recovery does not need to happen on remount.
1522 */
1523
1524int journal_flush(journal_t *journal)
1525{
1526        int err = 0;
1527        transaction_t *transaction = NULL;
1528
1529        spin_lock(&journal->j_state_lock);
1530
1531        /* Force everything buffered to the log... */
1532        if (journal->j_running_transaction) {
1533                transaction = journal->j_running_transaction;
1534                __log_start_commit(journal, transaction->t_tid);
1535        } else if (journal->j_committing_transaction)
1536                transaction = journal->j_committing_transaction;
1537
1538        /* Wait for the log commit to complete... */
1539        if (transaction) {
1540                tid_t tid = transaction->t_tid;
1541
1542                spin_unlock(&journal->j_state_lock);
1543                log_wait_commit(journal, tid);
1544        } else {
1545                spin_unlock(&journal->j_state_lock);
1546        }
1547
1548        /* ...and flush everything in the log out to disk. */
1549        spin_lock(&journal->j_list_lock);
1550        while (!err && journal->j_checkpoint_transactions != NULL) {
1551                spin_unlock(&journal->j_list_lock);
1552                mutex_lock(&journal->j_checkpoint_mutex);
1553                err = log_do_checkpoint(journal);
1554                mutex_unlock(&journal->j_checkpoint_mutex);
1555                spin_lock(&journal->j_list_lock);
1556        }
1557        spin_unlock(&journal->j_list_lock);
1558
1559        if (is_journal_aborted(journal))
1560                return -EIO;
1561
1562        mutex_lock(&journal->j_checkpoint_mutex);
1563        cleanup_journal_tail(journal);
1564
1565        /* Finally, mark the journal as really needing no recovery.
1566         * This sets s_start==0 in the underlying superblock, which is
1567         * the magic code for a fully-recovered superblock.  Any future
1568         * commits of data to the journal will restore the current
1569         * s_start value. */
1570        mark_journal_empty(journal);
1571        mutex_unlock(&journal->j_checkpoint_mutex);
1572        spin_lock(&journal->j_state_lock);
1573        J_ASSERT(!journal->j_running_transaction);
1574        J_ASSERT(!journal->j_committing_transaction);
1575        J_ASSERT(!journal->j_checkpoint_transactions);
1576        J_ASSERT(journal->j_head == journal->j_tail);
1577        J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1578        spin_unlock(&journal->j_state_lock);
1579        return 0;
1580}
1581
1582/**
1583 * int journal_wipe() - Wipe journal contents
1584 * @journal: Journal to act on.
1585 * @write: flag (see below)
1586 *
1587 * Wipe out all of the contents of a journal, safely.  This will produce
1588 * a warning if the journal contains any valid recovery information.
1589 * Must be called between journal_init_*() and journal_load().
1590 *
1591 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1592 * we merely suppress recovery.
1593 */
1594
1595int journal_wipe(journal_t *journal, int write)
1596{
1597        int err = 0;
1598
1599        J_ASSERT (!(journal->j_flags & JFS_LOADED));
1600
1601        err = load_superblock(journal);
1602        if (err)
1603                return err;
1604
1605        if (!journal->j_tail)
1606                goto no_recovery;
1607
1608        printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1609                write ? "Clearing" : "Ignoring");
1610
1611        err = journal_skip_recovery(journal);
1612        if (write) {
1613                /* Lock to make assertions happy... */
1614                mutex_lock(&journal->j_checkpoint_mutex);
1615                mark_journal_empty(journal);
1616                mutex_unlock(&journal->j_checkpoint_mutex);
1617        }
1618
1619 no_recovery:
1620        return err;
1621}
1622
1623/*
1624 * journal_dev_name: format a character string to describe on what
1625 * device this journal is present.
1626 */
1627
1628static const char *journal_dev_name(journal_t *journal, char *buffer)
1629{
1630        struct block_device *bdev;
1631
1632        if (journal->j_inode)
1633                bdev = journal->j_inode->i_sb->s_bdev;
1634        else
1635                bdev = journal->j_dev;
1636
1637        return bdevname(bdev, buffer);
1638}
1639
1640/*
1641 * Journal abort has very specific semantics, which we describe
1642 * for journal abort.
1643 *
1644 * Two internal function, which provide abort to te jbd layer
1645 * itself are here.
1646 */
1647
1648/*
1649 * Quick version for internal journal use (doesn't lock the journal).
1650 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1651 * and don't attempt to make any other journal updates.
1652 */
1653static void __journal_abort_hard(journal_t *journal)
1654{
1655        transaction_t *transaction;
1656        char b[BDEVNAME_SIZE];
1657
1658        if (journal->j_flags & JFS_ABORT)
1659                return;
1660
1661        printk(KERN_ERR "Aborting journal on device %s.\n",
1662                journal_dev_name(journal, b));
1663
1664        spin_lock(&journal->j_state_lock);
1665        journal->j_flags |= JFS_ABORT;
1666        transaction = journal->j_running_transaction;
1667        if (transaction)
1668                __log_start_commit(journal, transaction->t_tid);
1669        spin_unlock(&journal->j_state_lock);
1670}
1671
1672/* Soft abort: record the abort error status in the journal superblock,
1673 * but don't do any other IO. */
1674static void __journal_abort_soft (journal_t *journal, int errno)
1675{
1676        if (journal->j_flags & JFS_ABORT)
1677                return;
1678
1679        if (!journal->j_errno)
1680                journal->j_errno = errno;
1681
1682        __journal_abort_hard(journal);
1683
1684        if (errno)
1685                journal_update_sb_errno(journal);
1686}
1687
1688/**
1689 * void journal_abort () - Shutdown the journal immediately.
1690 * @journal: the journal to shutdown.
1691 * @errno:   an error number to record in the journal indicating
1692 *           the reason for the shutdown.
1693 *
1694 * Perform a complete, immediate shutdown of the ENTIRE
1695 * journal (not of a single transaction).  This operation cannot be
1696 * undone without closing and reopening the journal.
1697 *
1698 * The journal_abort function is intended to support higher level error
1699 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1700 * mode.
1701 *
1702 * Journal abort has very specific semantics.  Any existing dirty,
1703 * unjournaled buffers in the main filesystem will still be written to
1704 * disk by bdflush, but the journaling mechanism will be suspended
1705 * immediately and no further transaction commits will be honoured.
1706 *
1707 * Any dirty, journaled buffers will be written back to disk without
1708 * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1709 * filesystem, but we _do_ attempt to leave as much data as possible
1710 * behind for fsck to use for cleanup.
1711 *
1712 * Any attempt to get a new transaction handle on a journal which is in
1713 * ABORT state will just result in an -EROFS error return.  A
1714 * journal_stop on an existing handle will return -EIO if we have
1715 * entered abort state during the update.
1716 *
1717 * Recursive transactions are not disturbed by journal abort until the
1718 * final journal_stop, which will receive the -EIO error.
1719 *
1720 * Finally, the journal_abort call allows the caller to supply an errno
1721 * which will be recorded (if possible) in the journal superblock.  This
1722 * allows a client to record failure conditions in the middle of a
1723 * transaction without having to complete the transaction to record the
1724 * failure to disk.  ext3_error, for example, now uses this
1725 * functionality.
1726 *
1727 * Errors which originate from within the journaling layer will NOT
1728 * supply an errno; a null errno implies that absolutely no further
1729 * writes are done to the journal (unless there are any already in
1730 * progress).
1731 *
1732 */
1733
1734void journal_abort(journal_t *journal, int errno)
1735{
1736        __journal_abort_soft(journal, errno);
1737}
1738
1739/**
1740 * int journal_errno () - returns the journal's error state.
1741 * @journal: journal to examine.
1742 *
1743 * This is the errno numbet set with journal_abort(), the last
1744 * time the journal was mounted - if the journal was stopped
1745 * without calling abort this will be 0.
1746 *
1747 * If the journal has been aborted on this mount time -EROFS will
1748 * be returned.
1749 */
1750int journal_errno(journal_t *journal)
1751{
1752        int err;
1753
1754        spin_lock(&journal->j_state_lock);
1755        if (journal->j_flags & JFS_ABORT)
1756                err = -EROFS;
1757        else
1758                err = journal->j_errno;
1759        spin_unlock(&journal->j_state_lock);
1760        return err;
1761}
1762
1763/**
1764 * int journal_clear_err () - clears the journal's error state
1765 * @journal: journal to act on.
1766 *
1767 * An error must be cleared or Acked to take a FS out of readonly
1768 * mode.
1769 */
1770int journal_clear_err(journal_t *journal)
1771{
1772        int err = 0;
1773
1774        spin_lock(&journal->j_state_lock);
1775        if (journal->j_flags & JFS_ABORT)
1776                err = -EROFS;
1777        else
1778                journal->j_errno = 0;
1779        spin_unlock(&journal->j_state_lock);
1780        return err;
1781}
1782
1783/**
1784 * void journal_ack_err() - Ack journal err.
1785 * @journal: journal to act on.
1786 *
1787 * An error must be cleared or Acked to take a FS out of readonly
1788 * mode.
1789 */
1790void journal_ack_err(journal_t *journal)
1791{
1792        spin_lock(&journal->j_state_lock);
1793        if (journal->j_errno)
1794                journal->j_flags |= JFS_ACK_ERR;
1795        spin_unlock(&journal->j_state_lock);
1796}
1797
1798int journal_blocks_per_page(struct inode *inode)
1799{
1800        return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1801}
1802
1803/*
1804 * Journal_head storage management
1805 */
1806static struct kmem_cache *journal_head_cache;
1807#ifdef CONFIG_JBD_DEBUG
1808static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1809#endif
1810
1811static int journal_init_journal_head_cache(void)
1812{
1813        int retval;
1814
1815        J_ASSERT(journal_head_cache == NULL);
1816        journal_head_cache = kmem_cache_create("journal_head",
1817                                sizeof(struct journal_head),
1818                                0,              /* offset */
1819                                SLAB_TEMPORARY, /* flags */
1820                                NULL);          /* ctor */
1821        retval = 0;
1822        if (!journal_head_cache) {
1823                retval = -ENOMEM;
1824                printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1825        }
1826        return retval;
1827}
1828
1829static void journal_destroy_journal_head_cache(void)
1830{
1831        if (journal_head_cache) {
1832                kmem_cache_destroy(journal_head_cache);
1833                journal_head_cache = NULL;
1834        }
1835}
1836
1837/*
1838 * journal_head splicing and dicing
1839 */
1840static struct journal_head *journal_alloc_journal_head(void)
1841{
1842        struct journal_head *ret;
1843
1844#ifdef CONFIG_JBD_DEBUG
1845        atomic_inc(&nr_journal_heads);
1846#endif
1847        ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1848        if (ret == NULL) {
1849                jbd_debug(1, "out of memory for journal_head\n");
1850                printk_ratelimited(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1851                                   __func__);
1852
1853                while (ret == NULL) {
1854                        yield();
1855                        ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS);
1856                }
1857        }
1858        return ret;
1859}
1860
1861static void journal_free_journal_head(struct journal_head *jh)
1862{
1863#ifdef CONFIG_JBD_DEBUG
1864        atomic_dec(&nr_journal_heads);
1865        memset(jh, JBD_POISON_FREE, sizeof(*jh));
1866#endif
1867        kmem_cache_free(journal_head_cache, jh);
1868}
1869
1870/*
1871 * A journal_head is attached to a buffer_head whenever JBD has an
1872 * interest in the buffer.
1873 *
1874 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1875 * is set.  This bit is tested in core kernel code where we need to take
1876 * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
1877 * there.
1878 *
1879 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1880 *
1881 * When a buffer has its BH_JBD bit set it is immune from being released by
1882 * core kernel code, mainly via ->b_count.
1883 *
1884 * A journal_head is detached from its buffer_head when the journal_head's
1885 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
1886 * transaction (b_cp_transaction) hold their references to b_jcount.
1887 *
1888 * Various places in the kernel want to attach a journal_head to a buffer_head
1889 * _before_ attaching the journal_head to a transaction.  To protect the
1890 * journal_head in this situation, journal_add_journal_head elevates the
1891 * journal_head's b_jcount refcount by one.  The caller must call
1892 * journal_put_journal_head() to undo this.
1893 *
1894 * So the typical usage would be:
1895 *
1896 *      (Attach a journal_head if needed.  Increments b_jcount)
1897 *      struct journal_head *jh = journal_add_journal_head(bh);
1898 *      ...
1899 *      (Get another reference for transaction)
1900 *      journal_grab_journal_head(bh);
1901 *      jh->b_transaction = xxx;
1902 *      (Put original reference)
1903 *      journal_put_journal_head(jh);
1904 */
1905
1906/*
1907 * Give a buffer_head a journal_head.
1908 *
1909 * May sleep.
1910 */
1911struct journal_head *journal_add_journal_head(struct buffer_head *bh)
1912{
1913        struct journal_head *jh;
1914        struct journal_head *new_jh = NULL;
1915
1916repeat:
1917        if (!buffer_jbd(bh)) {
1918                new_jh = journal_alloc_journal_head();
1919                memset(new_jh, 0, sizeof(*new_jh));
1920        }
1921
1922        jbd_lock_bh_journal_head(bh);
1923        if (buffer_jbd(bh)) {
1924                jh = bh2jh(bh);
1925        } else {
1926                J_ASSERT_BH(bh,
1927                        (atomic_read(&bh->b_count) > 0) ||
1928                        (bh->b_page && bh->b_page->mapping));
1929
1930                if (!new_jh) {
1931                        jbd_unlock_bh_journal_head(bh);
1932                        goto repeat;
1933                }
1934
1935                jh = new_jh;
1936                new_jh = NULL;          /* We consumed it */
1937                set_buffer_jbd(bh);
1938                bh->b_private = jh;
1939                jh->b_bh = bh;
1940                get_bh(bh);
1941                BUFFER_TRACE(bh, "added journal_head");
1942        }
1943        jh->b_jcount++;
1944        jbd_unlock_bh_journal_head(bh);
1945        if (new_jh)
1946                journal_free_journal_head(new_jh);
1947        return bh->b_private;
1948}
1949
1950/*
1951 * Grab a ref against this buffer_head's journal_head.  If it ended up not
1952 * having a journal_head, return NULL
1953 */
1954struct journal_head *journal_grab_journal_head(struct buffer_head *bh)
1955{
1956        struct journal_head *jh = NULL;
1957
1958        jbd_lock_bh_journal_head(bh);
1959        if (buffer_jbd(bh)) {
1960                jh = bh2jh(bh);
1961                jh->b_jcount++;
1962        }
1963        jbd_unlock_bh_journal_head(bh);
1964        return jh;
1965}
1966
1967static void __journal_remove_journal_head(struct buffer_head *bh)
1968{
1969        struct journal_head *jh = bh2jh(bh);
1970
1971        J_ASSERT_JH(jh, jh->b_jcount >= 0);
1972        J_ASSERT_JH(jh, jh->b_transaction == NULL);
1973        J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1974        J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
1975        J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1976        J_ASSERT_BH(bh, buffer_jbd(bh));
1977        J_ASSERT_BH(bh, jh2bh(jh) == bh);
1978        BUFFER_TRACE(bh, "remove journal_head");
1979        if (jh->b_frozen_data) {
1980                printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
1981                jbd_free(jh->b_frozen_data, bh->b_size);
1982        }
1983        if (jh->b_committed_data) {
1984                printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
1985                jbd_free(jh->b_committed_data, bh->b_size);
1986        }
1987        bh->b_private = NULL;
1988        jh->b_bh = NULL;        /* debug, really */
1989        clear_buffer_jbd(bh);
1990        journal_free_journal_head(jh);
1991}
1992
1993/*
1994 * Drop a reference on the passed journal_head.  If it fell to zero then
1995 * release the journal_head from the buffer_head.
1996 */
1997void journal_put_journal_head(struct journal_head *jh)
1998{
1999        struct buffer_head *bh = jh2bh(jh);
2000
2001        jbd_lock_bh_journal_head(bh);
2002        J_ASSERT_JH(jh, jh->b_jcount > 0);
2003        --jh->b_jcount;
2004        if (!jh->b_jcount) {
2005                __journal_remove_journal_head(bh);
2006                jbd_unlock_bh_journal_head(bh);
2007                __brelse(bh);
2008        } else
2009                jbd_unlock_bh_journal_head(bh);
2010}
2011
2012/*
2013 * debugfs tunables
2014 */
2015#ifdef CONFIG_JBD_DEBUG
2016
2017u8 journal_enable_debug __read_mostly;
2018EXPORT_SYMBOL(journal_enable_debug);
2019
2020static struct dentry *jbd_debugfs_dir;
2021static struct dentry *jbd_debug;
2022
2023static void __init jbd_create_debugfs_entry(void)
2024{
2025        jbd_debugfs_dir = debugfs_create_dir("jbd", NULL);
2026        if (jbd_debugfs_dir)
2027                jbd_debug = debugfs_create_u8("jbd-debug", S_IRUGO | S_IWUSR,
2028                                               jbd_debugfs_dir,
2029                                               &journal_enable_debug);
2030}
2031
2032static void __exit jbd_remove_debugfs_entry(void)
2033{
2034        debugfs_remove(jbd_debug);
2035        debugfs_remove(jbd_debugfs_dir);
2036}
2037
2038#else
2039
2040static inline void jbd_create_debugfs_entry(void)
2041{
2042}
2043
2044static inline void jbd_remove_debugfs_entry(void)
2045{
2046}
2047
2048#endif
2049
2050struct kmem_cache *jbd_handle_cache;
2051
2052static int __init journal_init_handle_cache(void)
2053{
2054        jbd_handle_cache = kmem_cache_create("journal_handle",
2055                                sizeof(handle_t),
2056                                0,              /* offset */
2057                                SLAB_TEMPORARY, /* flags */
2058                                NULL);          /* ctor */
2059        if (jbd_handle_cache == NULL) {
2060                printk(KERN_EMERG "JBD: failed to create handle cache\n");
2061                return -ENOMEM;
2062        }
2063        return 0;
2064}
2065
2066static void journal_destroy_handle_cache(void)
2067{
2068        if (jbd_handle_cache)
2069                kmem_cache_destroy(jbd_handle_cache);
2070}
2071
2072/*
2073 * Module startup and shutdown
2074 */
2075
2076static int __init journal_init_caches(void)
2077{
2078        int ret;
2079
2080        ret = journal_init_revoke_caches();
2081        if (ret == 0)
2082                ret = journal_init_journal_head_cache();
2083        if (ret == 0)
2084                ret = journal_init_handle_cache();
2085        return ret;
2086}
2087
2088static void journal_destroy_caches(void)
2089{
2090        journal_destroy_revoke_caches();
2091        journal_destroy_journal_head_cache();
2092        journal_destroy_handle_cache();
2093}
2094
2095static int __init journal_init(void)
2096{
2097        int ret;
2098
2099        BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2100
2101        ret = journal_init_caches();
2102        if (ret != 0)
2103                journal_destroy_caches();
2104        jbd_create_debugfs_entry();
2105        return ret;
2106}
2107
2108static void __exit journal_exit(void)
2109{
2110#ifdef CONFIG_JBD_DEBUG
2111        int n = atomic_read(&nr_journal_heads);
2112        if (n)
2113                printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2114#endif
2115        jbd_remove_debugfs_entry();
2116        journal_destroy_caches();
2117}
2118
2119MODULE_LICENSE("GPL");
2120module_init(journal_init);
2121module_exit(journal_exit);
2122
2123
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.