Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / fs / jbd / journal.c
index 3c95df4..7f96b5c 100644 (file)
 #include <linux/mm.h>
 #include <linux/suspend.h>
 #include <linux/pagemap.h>
-#include <asm/uaccess.h>
+#include <linux/kthread.h>
 #include <linux/proc_fs.h>
 
+#include <asm/uaccess.h>
+#include <asm/page.h>
+
 EXPORT_SYMBOL(journal_start);
 EXPORT_SYMBOL(journal_restart);
 EXPORT_SYMBOL(journal_extend);
@@ -54,7 +57,6 @@ EXPORT_SYMBOL(journal_sync_buffer);
 #endif
 EXPORT_SYMBOL(journal_flush);
 EXPORT_SYMBOL(journal_revoke);
-EXPORT_SYMBOL(journal_callback_set);
 
 EXPORT_SYMBOL(journal_init_dev);
 EXPORT_SYMBOL(journal_init_inode);
@@ -65,7 +67,6 @@ EXPORT_SYMBOL(journal_set_features);
 EXPORT_SYMBOL(journal_create);
 EXPORT_SYMBOL(journal_load);
 EXPORT_SYMBOL(journal_destroy);
-EXPORT_SYMBOL(journal_recover);
 EXPORT_SYMBOL(journal_update_superblock);
 EXPORT_SYMBOL(journal_abort);
 EXPORT_SYMBOL(journal_errno);
@@ -73,14 +74,15 @@ EXPORT_SYMBOL(journal_ack_err);
 EXPORT_SYMBOL(journal_clear_err);
 EXPORT_SYMBOL(log_wait_commit);
 EXPORT_SYMBOL(journal_start_commit);
+EXPORT_SYMBOL(journal_force_commit_nested);
 EXPORT_SYMBOL(journal_wipe);
 EXPORT_SYMBOL(journal_blocks_per_page);
 EXPORT_SYMBOL(journal_invalidatepage);
 EXPORT_SYMBOL(journal_try_to_free_buffers);
-EXPORT_SYMBOL(journal_bmap);
 EXPORT_SYMBOL(journal_force_commit);
 
 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
+static void __journal_abort_soft (journal_t *journal, int errno);
 
 /*
  * Helper function used to manage commit timeouts
@@ -93,16 +95,6 @@ static void commit_timeout(unsigned long __data)
        wake_up_process(p);
 }
 
-/* Static check for data structure consistency.  There's no code
- * invoked --- we'll just get a linker failure if things aren't right.
- */
-void __journal_internal_check(void)
-{
-       extern void journal_bad_superblock_size(void);
-       if (sizeof(struct journal_superblock_s) != 1024)
-               journal_bad_superblock_size();
-}
-
 /*
  * kjournald: The main thread function used to manage a logging device
  * journal.
@@ -119,24 +111,17 @@ void __journal_internal_check(void)
  *    known as checkpointing, and this thread is responsible for that job.
  */
 
-journal_t *current_journal;            // AKPM: debug
-
-int kjournald(void *arg)
+static int kjournald(void *arg)
 {
-       journal_t *journal = (journal_t *) arg;
+       journal_t *journal = arg;
        transaction_t *transaction;
-       struct timer_list timer;
 
-       current_journal = journal;
-
-       daemonize("kjournald");
-
-       /* Set up an interval timer which can be used to trigger a
-           commit wakeup after the commit interval expires */
-       init_timer(&timer);
-       timer.data = (unsigned long) current;
-       timer.function = commit_timeout;
-       journal->j_commit_timer = &timer;
+       /*
+        * Set up an interval timer which can be used to trigger a commit wakeup
+        * after the commit interval expires
+        */
+       setup_timer(&journal->j_commit_timer, commit_timeout,
+                       (unsigned long)current);
 
        /* Record that the journal thread is running */
        journal->j_task = current;
@@ -151,20 +136,23 @@ int kjournald(void *arg)
        spin_lock(&journal->j_state_lock);
 
 loop:
+       if (journal->j_flags & JFS_UNMOUNT)
+               goto end_loop;
+
        jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
                journal->j_commit_sequence, journal->j_commit_request);
 
        if (journal->j_commit_sequence != journal->j_commit_request) {
                jbd_debug(1, "OK, requests differ\n");
                spin_unlock(&journal->j_state_lock);
-               del_timer_sync(journal->j_commit_timer);
+               del_timer_sync(&journal->j_commit_timer);
                journal_commit_transaction(journal);
                spin_lock(&journal->j_state_lock);
-               goto end_loop;
+               goto loop;
        }
 
        wake_up(&journal->j_wait_done_commit);
-       if (current->flags & PF_FREEZE) {
+       if (freezing(current)) {
                /*
                 * The simpler the better. Flushing journal isn't a
                 * good idea, because that depends on threads that may
@@ -172,7 +160,7 @@ loop:
                 */
                jbd_debug(1, "Now suspending kjournald\n");
                spin_unlock(&journal->j_state_lock);
-               refrigerator(PF_FREEZE);
+               refrigerator();
                spin_lock(&journal->j_state_lock);
        } else {
                /*
@@ -190,6 +178,8 @@ loop:
                if (transaction && time_after_eq(jiffies,
                                                transaction->t_expires))
                        should_sleep = 0;
+               if (journal->j_flags & JFS_UNMOUNT)
+                       should_sleep = 0;
                if (should_sleep) {
                        spin_unlock(&journal->j_state_lock);
                        schedule();
@@ -208,12 +198,11 @@ loop:
                journal->j_commit_request = transaction->t_tid;
                jbd_debug(1, "woke because of timeout\n");
        }
-end_loop:
-       if (!(journal->j_flags & JFS_UNMOUNT))
-               goto loop;
+       goto loop;
 
+end_loop:
        spin_unlock(&journal->j_state_lock);
-       del_timer_sync(journal->j_commit_timer);
+       del_timer_sync(&journal->j_commit_timer);
        journal->j_task = NULL;
        wake_up(&journal->j_wait_done_commit);
        jbd_debug(1, "Journal thread exiting.\n");
@@ -222,7 +211,7 @@ end_loop:
 
 static void journal_start_thread(journal_t *journal)
 {
-       kernel_thread(kjournald, journal, CLONE_VM|CLONE_FS|CLONE_FILES);
+       kthread_run(kjournald, journal, "kjournald");
        wait_event(journal->j_wait_done_commit, journal->j_task != 0);
 }
 
@@ -324,8 +313,8 @@ repeat:
        /*
         * Check for escaping
         */
-       if (*((unsigned int *)(mapped_data + new_offset)) ==
-                               htonl(JFS_MAGIC_NUMBER)) {
+       if (*((__be32 *)(mapped_data + new_offset)) ==
+                               cpu_to_be32(JFS_MAGIC_NUMBER)) {
                need_copy_out = 1;
                do_escape = 1;
        }
@@ -464,6 +453,39 @@ int log_start_commit(journal_t *journal, tid_t tid)
        return ret;
 }
 
+/*
+ * Force and wait upon a commit if the calling process is not within
+ * transaction.  This is used for forcing out undo-protected data which contains
+ * bitmaps, when the fs is running out of space.
+ *
+ * We can only force the running transaction if we don't have an active handle;
+ * otherwise, we will deadlock.
+ *
+ * Returns true if a transaction was started.
+ */
+int journal_force_commit_nested(journal_t *journal)
+{
+       transaction_t *transaction = NULL;
+       tid_t tid;
+
+       spin_lock(&journal->j_state_lock);
+       if (journal->j_running_transaction && !current->journal_info) {
+               transaction = journal->j_running_transaction;
+               __log_start_commit(journal, transaction->t_tid);
+       } else if (journal->j_committing_transaction)
+               transaction = journal->j_committing_transaction;
+
+       if (!transaction) {
+               spin_unlock(&journal->j_state_lock);
+               return 0;       /* Nothing to retry */
+       }
+
+       tid = transaction->t_tid;
+       spin_unlock(&journal->j_state_lock);
+       log_wait_commit(journal, tid);
+       return 1;
+}
+
 /*
  * Start a commit of the current running transaction (if any).  Returns true
  * if a transaction was started, and fills its tid in at *ptid
@@ -585,9 +607,13 @@ int journal_bmap(journal_t *journal, unsigned long blocknr,
  * We play buffer_head aliasing tricks to write data/metadata blocks to
  * the journal without copying their contents, but for journal
  * descriptor blocks we do need to generate bona fide buffers.
+ *
+ * After the caller of journal_get_descriptor_buffer() has finished modifying
+ * the buffer's contents they really should run flush_dcache_page(bh->b_page).
+ * But we don't bother doing that, so there will be coherency problems with
+ * mmaps of blockdevs which hold live JBD-controlled filesystems.
  */
-
-struct journal_head * journal_get_descriptor_buffer(journal_t *journal)
+struct journal_head *journal_get_descriptor_buffer(journal_t *journal)
 {
        struct buffer_head *bh;
        unsigned long blocknr;
@@ -599,8 +625,10 @@ struct journal_head * journal_get_descriptor_buffer(journal_t *journal)
                return NULL;
 
        bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
+       lock_buffer(bh);
        memset(bh->b_data, 0, journal->j_blocksize);
-       bh->b_state |= (1 << BH_Dirty);
+       set_buffer_uptodate(bh);
+       unlock_buffer(bh);
        BUFFER_TRACE(bh, "return this buffer");
        return journal_add_journal_head(bh);
 }
@@ -630,8 +658,8 @@ static journal_t * journal_init_common (void)
        init_waitqueue_head(&journal->j_wait_checkpoint);
        init_waitqueue_head(&journal->j_wait_commit);
        init_waitqueue_head(&journal->j_wait_updates);
-       init_MUTEX(&journal->j_barrier);
-       init_MUTEX(&journal->j_checkpoint_sem);
+       mutex_init(&journal->j_barrier);
+       mutex_init(&journal->j_checkpoint_mutex);
        spin_lock_init(&journal->j_revoke_lock);
        spin_lock_init(&journal->j_list_lock);
        spin_lock_init(&journal->j_state_lock);
@@ -680,6 +708,7 @@ journal_t * journal_init_dev(struct block_device *bdev,
 {
        journal_t *journal = journal_init_common();
        struct buffer_head *bh;
+       int n;
 
        if (!journal)
                return NULL;
@@ -695,6 +724,17 @@ journal_t * journal_init_dev(struct block_device *bdev,
        journal->j_sb_buffer = bh;
        journal->j_superblock = (journal_superblock_t *)bh->b_data;
 
+       /* journal descriptor can store up to n blocks -bzzz */
+       n = journal->j_blocksize / sizeof(journal_block_tag_t);
+       journal->j_wbufsize = n;
+       journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
+       if (!journal->j_wbuf) {
+               printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
+                       __FUNCTION__);
+               kfree(journal);
+               journal = NULL;
+       }
+
        return journal;
 }
  
@@ -711,6 +751,7 @@ journal_t * journal_init_inode (struct inode *inode)
        struct buffer_head *bh;
        journal_t *journal = journal_init_common();
        int err;
+       int n;
        unsigned long blocknr;
 
        if (!journal)
@@ -727,6 +768,17 @@ journal_t * journal_init_inode (struct inode *inode)
        journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
        journal->j_blocksize = inode->i_sb->s_blocksize;
 
+       /* journal descriptor can store up to n blocks -bzzz */
+       n = journal->j_blocksize / sizeof(journal_block_tag_t);
+       journal->j_wbufsize = n;
+       journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
+       if (!journal->j_wbuf) {
+               printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
+                       __FUNCTION__);
+               kfree(journal);
+               return NULL;
+       }
+
        err = journal_bmap(journal, 0, &blocknr);
        /* If that failed, give up */
        if (err) {
@@ -768,8 +820,8 @@ static int journal_reset(journal_t *journal)
        journal_superblock_t *sb = journal->j_superblock;
        unsigned int first, last;
 
-       first = ntohl(sb->s_first);
-       last = ntohl(sb->s_maxlen);
+       first = be32_to_cpu(sb->s_first);
+       last = be32_to_cpu(sb->s_maxlen);
 
        journal->j_first = first;
        journal->j_last = last;
@@ -846,12 +898,12 @@ int journal_create(journal_t *journal)
        /* OK, fill in the initial static fields in the new superblock */
        sb = journal->j_superblock;
 
-       sb->s_header.h_magic     = htonl(JFS_MAGIC_NUMBER);
-       sb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
+       sb->s_header.h_magic     = cpu_to_be32(JFS_MAGIC_NUMBER);
+       sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2);
 
-       sb->s_blocksize = htonl(journal->j_blocksize);
-       sb->s_maxlen    = htonl(journal->j_maxlen);
-       sb->s_first     = htonl(1);
+       sb->s_blocksize = cpu_to_be32(journal->j_blocksize);
+       sb->s_maxlen    = cpu_to_be32(journal->j_maxlen);
+       sb->s_first     = cpu_to_be32(1);
 
        journal->j_transaction_sequence = 1;
 
@@ -894,9 +946,9 @@ void journal_update_superblock(journal_t *journal, int wait)
        jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
                  journal->j_tail, journal->j_tail_sequence, journal->j_errno);
 
-       sb->s_sequence = htonl(journal->j_tail_sequence);
-       sb->s_start    = htonl(journal->j_tail);
-       sb->s_errno    = htonl(journal->j_errno);
+       sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
+       sb->s_start    = cpu_to_be32(journal->j_tail);
+       sb->s_errno    = cpu_to_be32(journal->j_errno);
        spin_unlock(&journal->j_state_lock);
 
        BUFFER_TRACE(bh, "marking dirty");
@@ -904,7 +956,7 @@ void journal_update_superblock(journal_t *journal, int wait)
        if (wait)
                sync_dirty_buffer(bh);
        else
-               ll_rw_block(WRITE, 1, &bh);
+               ll_rw_block(SWRITE, 1, &bh);
 
 out:
        /* If we have just flushed the log (by marking s_start==0), then
@@ -947,13 +999,13 @@ static int journal_get_superblock(journal_t *journal)
 
        err = -EINVAL;
 
-       if (sb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
-           sb->s_blocksize != htonl(journal->j_blocksize)) {
+       if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) ||
+           sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
                printk(KERN_WARNING "JBD: no valid journal superblock found\n");
                goto out;
        }
 
-       switch(ntohl(sb->s_header.h_blocktype)) {
+       switch(be32_to_cpu(sb->s_header.h_blocktype)) {
        case JFS_SUPERBLOCK_V1:
                journal->j_format_version = 1;
                break;
@@ -965,9 +1017,9 @@ static int journal_get_superblock(journal_t *journal)
                goto out;
        }
 
-       if (ntohl(sb->s_maxlen) < journal->j_maxlen)
-               journal->j_maxlen = ntohl(sb->s_maxlen);
-       else if (ntohl(sb->s_maxlen) > journal->j_maxlen) {
+       if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
+               journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
+       else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
                printk (KERN_WARNING "JBD: journal file too short\n");
                goto out;
        }
@@ -995,11 +1047,11 @@ static int load_superblock(journal_t *journal)
 
        sb = journal->j_superblock;
 
-       journal->j_tail_sequence = ntohl(sb->s_sequence);
-       journal->j_tail = ntohl(sb->s_start);
-       journal->j_first = ntohl(sb->s_first);
-       journal->j_last = ntohl(sb->s_maxlen);
-       journal->j_errno = ntohl(sb->s_errno);
+       journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
+       journal->j_tail = be32_to_cpu(sb->s_start);
+       journal->j_first = be32_to_cpu(sb->s_first);
+       journal->j_last = be32_to_cpu(sb->s_maxlen);
+       journal->j_errno = be32_to_cpu(sb->s_errno);
 
        return 0;
 }
@@ -1100,12 +1152,17 @@ void journal_destroy(journal_t *journal)
                iput(journal->j_inode);
        if (journal->j_revoke)
                journal_destroy_revoke(journal);
+       kfree(journal->j_wbuf);
        kfree(journal);
 }
 
 
 /**
  *int journal_check_used_features () - Check if features specified are used.
+ * @journal: Journal to check.
+ * @compat: bitmask of compatible features
+ * @ro: bitmask of features that force read-only mount
+ * @incompat: bitmask of incompatible features
  * 
  * Check whether the journal uses all of a given set of
  * features.  Return true (non-zero) if it does. 
@@ -1133,6 +1190,10 @@ int journal_check_used_features (journal_t *journal, unsigned long compat,
 
 /**
  * int journal_check_available_features() - Check feature set in journalling layer
+ * @journal: Journal to check.
+ * @compat: bitmask of compatible features
+ * @ro: bitmask of features that force read-only mount
+ * @incompat: bitmask of incompatible features
  * 
  * Check whether the journaling code supports the use of
  * all of a given set of features on this journal.  Return true
@@ -1165,6 +1226,10 @@ int journal_check_available_features (journal_t *journal, unsigned long compat,
 
 /**
  * int journal_set_features () - Mark a given journal feature in the superblock
+ * @journal: Journal to act on.
+ * @compat: bitmask of compatible features
+ * @ro: bitmask of features that force read-only mount
+ * @incompat: bitmask of incompatible features
  *
  * Mark a given journal feature as present on the
  * superblock.  Returns true if the requested features could be set. 
@@ -1197,6 +1262,7 @@ int journal_set_features (journal_t *journal, unsigned long compat,
 
 /**
  * int journal_update_format () - Update on-disk journal structure.
+ * @journal: Journal to act on.
  *
  * Given an initialised but unloaded journal struct, poke about in the
  * on-disk structure to update it to the most recent supported version.
@@ -1212,7 +1278,7 @@ int journal_update_format (journal_t *journal)
 
        sb = journal->j_superblock;
 
-       switch (ntohl(sb->s_header.h_blocktype)) {
+       switch (be32_to_cpu(sb->s_header.h_blocktype)) {
        case JFS_SUPERBLOCK_V2:
                return 0;
        case JFS_SUPERBLOCK_V1:
@@ -1234,7 +1300,7 @@ static int journal_convert_superblock_v1(journal_t *journal,
 
        /* Pre-initialise new fields to zero */
        offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
-       blocksize = ntohl(sb->s_blocksize);
+       blocksize = be32_to_cpu(sb->s_blocksize);
        memset(&sb->s_feature_compat, 0, blocksize-offset);
 
        sb->s_nr_users = cpu_to_be32(1);
@@ -1360,7 +1426,7 @@ int journal_wipe(journal_t *journal, int write)
  * device this journal is present.
  */
 
-const char *journal_dev_name(journal_t *journal, char *buffer)
+static const char *journal_dev_name(journal_t *journal, char *buffer)
 {
        struct block_device *bdev;
 
@@ -1406,7 +1472,7 @@ void __journal_abort_hard(journal_t *journal)
 
 /* Soft abort: record the abort error status in the journal superblock,
  * but don't do any other IO. */
-void __journal_abort_soft (journal_t *journal, int errno)
+static void __journal_abort_soft (journal_t *journal, int errno)
 {
        if (journal->j_flags & JFS_ABORT)
                return;
@@ -1497,6 +1563,7 @@ int journal_errno(journal_t *journal)
 
 /** 
  * int journal_clear_err () - clears the journal's error state
+ * @journal: journal to act on.
  *
  * An error must be cleared or Acked to take a FS out of readonly
  * mode.
@@ -1516,6 +1583,7 @@ int journal_clear_err(journal_t *journal)
 
 /** 
  * void journal_ack_err() - Ack journal err.
+ * @journal: journal to act on.
  *
  * An error must be cleared or Acked to take a FS out of readonly
  * mode.
@@ -1534,14 +1602,10 @@ int journal_blocks_per_page(struct inode *inode)
 }
 
 /*
- * Simple support for retying memory allocations.  Introduced to help to
- * debug different VM deadlock avoidance strategies. 
- */
-/*
- * Simple support for retying memory allocations.  Introduced to help to
+ * Simple support for retrying memory allocations.  Introduced to help to
  * debug different VM deadlock avoidance strategies. 
  */
-void * __jbd_kmalloc (const char *where, size_t size, int flags, int retry)
+void * __jbd_kmalloc (const char *where, size_t size, gfp_t flags, int retry)
 {
        return kmalloc(size, flags | (retry ? __GFP_NOFAIL : 0));
 }
@@ -1577,7 +1641,7 @@ static void journal_destroy_journal_head_cache(void)
 {
        J_ASSERT(journal_head_cache != NULL);
        kmem_cache_destroy(journal_head_cache);
-       journal_head_cache = 0;
+       journal_head_cache = NULL;
 }
 
 /*
@@ -1726,6 +1790,7 @@ static void __journal_remove_journal_head(struct buffer_head *bh)
                if (jh->b_transaction == NULL &&
                                jh->b_next_transaction == NULL &&
                                jh->b_cp_transaction == NULL) {
+                       J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
                        J_ASSERT_BH(bh, buffer_jbd(bh));
                        J_ASSERT_BH(bh, jh2bh(jh) == bh);
                        BUFFER_TRACE(bh, "remove journal_head");
@@ -1802,7 +1867,7 @@ EXPORT_SYMBOL(journal_enable_debug);
 
 static struct proc_dir_entry *proc_jbd_debug;
 
-int read_jbd_debug(char *page, char **start, off_t off,
+static int read_jbd_debug(char *page, char **start, off_t off,
                          int count, int *eof, void *data)
 {
        int ret;
@@ -1812,7 +1877,7 @@ int read_jbd_debug(char *page, char **start, off_t off,
        return ret;
 }
 
-int write_jbd_debug(struct file *file, const char __user *buffer,
+static int write_jbd_debug(struct file *file, const char __user *buffer,
                           unsigned long count, void *data)
 {
        char buf[32];
@@ -1901,6 +1966,14 @@ static int __init journal_init(void)
 {
        int ret;
 
+/* Static check for data structure consistency.  There's no code
+ * invoked --- we'll just get a linker failure if things aren't right.
+ */
+       extern void journal_bad_superblock_size(void);
+       if (sizeof(struct journal_superblock_s) != 1024)
+               journal_bad_superblock_size();
+
+
        ret = journal_init_caches();
        if (ret != 0)
                journal_destroy_caches();