Btrfs: find ideal block group for caching
[safe/jmp/linux-2.6] / fs / jbd2 / journal.c
index 5814410..e378cb3 100644 (file)
 #include <linux/debugfs.h>
 #include <linux/seq_file.h>
 #include <linux/math64.h>
+#include <linux/hash.h>
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/jbd2.h>
 
 #include <asm/uaccess.h>
 #include <asm/page.h>
@@ -293,6 +297,7 @@ int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
        unsigned int new_offset;
        struct buffer_head *bh_in = jh2bh(jh_in);
        struct jbd2_buffer_trigger_type *triggers;
+       journal_t *journal = transaction->t_journal;
 
        /*
         * The buffer really shouldn't be locked: only the current committing
@@ -306,6 +311,11 @@ int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
        J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
 
        new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
+       /* keep subsequent assertions sane */
+       new_bh->b_state = 0;
+       init_buffer(new_bh, NULL, NULL);
+       atomic_set(&new_bh->b_count, 1);
+       new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
 
        /*
         * If a new transaction has already done a buffer copy-out, then
@@ -384,14 +394,6 @@ repeat:
                kunmap_atomic(mapped_data, KM_USER0);
        }
 
-       /* keep subsequent assertions sane */
-       new_bh->b_state = 0;
-       init_buffer(new_bh, NULL, NULL);
-       atomic_set(&new_bh->b_count, 1);
-       jbd_unlock_bh_state(bh_in);
-
-       new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
-
        set_bh_page(new_bh, new_page, new_offset);
        new_jh->b_transaction = NULL;
        new_bh->b_size = jh2bh(jh_in)->b_size;
@@ -408,7 +410,11 @@ repeat:
         * copying is moved to the transaction's shadow queue.
         */
        JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
-       jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
+       spin_lock(&journal->j_list_lock);
+       __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
+       spin_unlock(&journal->j_list_lock);
+       jbd_unlock_bh_state(bh_in);
+
        JBUFFER_TRACE(new_jh, "file as BJ_IO");
        jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
 
@@ -1781,7 +1787,7 @@ int jbd2_journal_wipe(journal_t *journal, int write)
  * Journal abort has very specific semantics, which we describe
  * for journal abort.
  *
- * Two internal function, which provide abort to te jbd layer
+ * Two internal functions, which provide abort to the jbd layer
  * itself are here.
  */
 
@@ -1879,7 +1885,7 @@ void jbd2_journal_abort(journal_t *journal, int errno)
  * int jbd2_journal_errno () - returns the journal's error state.
  * @journal: journal to examine.
  *
- * This is the errno numbet set with jbd2_journal_abort(), the last
+ * This is the errno number set with jbd2_journal_abort(), the last
  * time the journal was mounted - if the journal was stopped
  * without calling abort this will be 0.
  *
@@ -1903,7 +1909,7 @@ int jbd2_journal_errno(journal_t *journal)
  * int jbd2_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
+ * An error must be cleared or acked to take a FS out of readonly
  * mode.
  */
 int jbd2_journal_clear_err(journal_t *journal)
@@ -1923,7 +1929,7 @@ int jbd2_journal_clear_err(journal_t *journal)
  * void jbd2_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
+ * An error must be cleared or acked to take a FS out of readonly
  * mode.
  */
 void jbd2_journal_ack_err(journal_t *journal)
@@ -2377,6 +2383,72 @@ static void __exit journal_exit(void)
        jbd2_journal_destroy_caches();
 }
 
+/* 
+ * jbd2_dev_to_name is a utility function used by the jbd2 and ext4 
+ * tracing infrastructure to map a dev_t to a device name.
+ *
+ * The caller should use rcu_read_lock() in order to make sure the
+ * device name stays valid until its done with it.  We use
+ * rcu_read_lock() as well to make sure we're safe in case the caller
+ * gets sloppy, and because rcu_read_lock() is cheap and can be safely
+ * nested.
+ */
+struct devname_cache {
+       struct rcu_head rcu;
+       dev_t           device;
+       char            devname[BDEVNAME_SIZE];
+};
+#define CACHE_SIZE_BITS 6
+static struct devname_cache *devcache[1 << CACHE_SIZE_BITS];
+static DEFINE_SPINLOCK(devname_cache_lock);
+
+static void free_devcache(struct rcu_head *rcu)
+{
+       kfree(rcu);
+}
+
+const char *jbd2_dev_to_name(dev_t device)
+{
+       int     i = hash_32(device, CACHE_SIZE_BITS);
+       char    *ret;
+       struct block_device *bd;
+       static struct devname_cache *new_dev;
+
+       rcu_read_lock();
+       if (devcache[i] && devcache[i]->device == device) {
+               ret = devcache[i]->devname;
+               rcu_read_unlock();
+               return ret;
+       }
+       rcu_read_unlock();
+
+       new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
+       if (!new_dev)
+               return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
+       spin_lock(&devname_cache_lock);
+       if (devcache[i]) {
+               if (devcache[i]->device == device) {
+                       kfree(new_dev);
+                       ret = devcache[i]->devname;
+                       spin_unlock(&devname_cache_lock);
+                       return ret;
+               }
+               call_rcu(&devcache[i]->rcu, free_devcache);
+       }
+       devcache[i] = new_dev;
+       devcache[i]->device = device;
+       bd = bdget(device);
+       if (bd) {
+               bdevname(bd, devcache[i]->devname);
+               bdput(bd);
+       } else
+               __bdevname(device, devcache[i]->devname);
+       ret = devcache[i]->devname;
+       spin_unlock(&devname_cache_lock);
+       return ret;
+}
+EXPORT_SYMBOL(jbd2_dev_to_name);
+
 MODULE_LICENSE("GPL");
 module_init(journal_init);
 module_exit(journal_exit);