Merge commit 'v2.6.30' into for-2.6.31
[safe/jmp/linux-2.6] / fs / jbd / revoke.c
index ad2eacf..da6cd9b 100644 (file)
  *                     need do nothing.
  * RevokeValid set, Revoked set:
  *                     buffer has been revoked.
+ *
+ * Locking rules:
+ * We keep two hash tables of revoke records. One hashtable belongs to the
+ * running transaction (is pointed to by journal->j_revoke), the other one
+ * belongs to the committing transaction. Accesses to the second hash table
+ * happen only from the kjournald and no other thread touches this table.  Also
+ * journal_switch_revoke_table() which switches which hashtable belongs to the
+ * running and which to the committing transaction is called only from
+ * kjournald. Therefore we need no locks when accessing the hashtable belonging
+ * to the committing transaction.
+ *
+ * All users operating on the hash table belonging to the running transaction
+ * have a handle to the transaction. Therefore they are safe from kjournald
+ * switching hash tables under them. For operations on the lists of entries in
+ * the hash table j_revoke_lock is used.
+ *
+ * Finally, also replay code uses the hash tables but at this moment noone else
+ * can touch them (filesystem isn't mounted yet) and hence no locking is
+ * needed.
  */
 
 #ifndef __KERNEL__
@@ -67,6 +86,7 @@
 #include <linux/slab.h>
 #include <linux/list.h>
 #include <linux/init.h>
+#include <linux/bio.h>
 #endif
 #include <linux/log2.h>
 
@@ -99,8 +119,8 @@ struct jbd_revoke_table_s
 #ifdef __KERNEL__
 static void write_one_revoke_record(journal_t *, transaction_t *,
                                    struct journal_head **, int *,
-                                   struct jbd_revoke_record_s *);
-static void flush_descriptor(journal_t *, struct journal_head *, int);
+                                   struct jbd_revoke_record_s *, int);
+static void flush_descriptor(journal_t *, struct journal_head *, int, int);
 #endif
 
 /* Utility functions to maintain the revoke table */
@@ -138,7 +158,7 @@ repeat:
 oom:
        if (!journal_oom_retry)
                return -ENOMEM;
-       jbd_debug(1, "ENOMEM in %s, retrying\n", __FUNCTION__);
+       jbd_debug(1, "ENOMEM in %s, retrying\n", __func__);
        yield();
        goto repeat;
 }
@@ -166,138 +186,123 @@ static struct jbd_revoke_record_s *find_revoke_record(journal_t *journal,
        return NULL;
 }
 
+void journal_destroy_revoke_caches(void)
+{
+       if (revoke_record_cache) {
+               kmem_cache_destroy(revoke_record_cache);
+               revoke_record_cache = NULL;
+       }
+       if (revoke_table_cache) {
+               kmem_cache_destroy(revoke_table_cache);
+               revoke_table_cache = NULL;
+       }
+}
+
 int __init journal_init_revoke_caches(void)
 {
+       J_ASSERT(!revoke_record_cache);
+       J_ASSERT(!revoke_table_cache);
+
        revoke_record_cache = kmem_cache_create("revoke_record",
                                           sizeof(struct jbd_revoke_record_s),
                                           0,
                                           SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
                                           NULL);
-       if (revoke_record_cache == 0)
-               return -ENOMEM;
+       if (!revoke_record_cache)
+               goto record_cache_failure;
 
        revoke_table_cache = kmem_cache_create("revoke_table",
                                           sizeof(struct jbd_revoke_table_s),
                                           0, SLAB_TEMPORARY, NULL);
-       if (revoke_table_cache == 0) {
-               kmem_cache_destroy(revoke_record_cache);
-               revoke_record_cache = NULL;
-               return -ENOMEM;
-       }
+       if (!revoke_table_cache)
+               goto table_cache_failure;
+
        return 0;
-}
 
-void journal_destroy_revoke_caches(void)
-{
-       kmem_cache_destroy(revoke_record_cache);
-       revoke_record_cache = NULL;
-       kmem_cache_destroy(revoke_table_cache);
-       revoke_table_cache = NULL;
+table_cache_failure:
+       journal_destroy_revoke_caches();
+record_cache_failure:
+       return -ENOMEM;
 }
 
-/* Initialise the revoke table for a given journal to a given size. */
-
-int journal_init_revoke(journal_t *journal, int hash_size)
+static struct jbd_revoke_table_s *journal_init_revoke_table(int hash_size)
 {
-       int shift, tmp;
+       int shift = 0;
+       int tmp = hash_size;
+       struct jbd_revoke_table_s *table;
 
-       J_ASSERT (journal->j_revoke_table[0] == NULL);
+       table = kmem_cache_alloc(revoke_table_cache, GFP_KERNEL);
+       if (!table)
+               goto out;
 
-       shift = 0;
-       tmp = hash_size;
        while((tmp >>= 1UL) != 0UL)
                shift++;
 
-       journal->j_revoke_table[0] = kmem_cache_alloc(revoke_table_cache, GFP_KERNEL);
-       if (!journal->j_revoke_table[0])
-               return -ENOMEM;
-       journal->j_revoke = journal->j_revoke_table[0];
-
-       /* Check that the hash_size is a power of two */
-       J_ASSERT(is_power_of_2(hash_size));
-
-       journal->j_revoke->hash_size = hash_size;
-
-       journal->j_revoke->hash_shift = shift;
-
-       journal->j_revoke->hash_table =
+       table->hash_size = hash_size;
+       table->hash_shift = shift;
+       table->hash_table =
                kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
-       if (!journal->j_revoke->hash_table) {
-               kmem_cache_free(revoke_table_cache, journal->j_revoke_table[0]);
-               journal->j_revoke = NULL;
-               return -ENOMEM;
+       if (!table->hash_table) {
+               kmem_cache_free(revoke_table_cache, table);
+               table = NULL;
+               goto out;
        }
 
        for (tmp = 0; tmp < hash_size; tmp++)
-               INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]);
+               INIT_LIST_HEAD(&table->hash_table[tmp]);
 
-       journal->j_revoke_table[1] = kmem_cache_alloc(revoke_table_cache, GFP_KERNEL);
-       if (!journal->j_revoke_table[1]) {
-               kfree(journal->j_revoke_table[0]->hash_table);
-               kmem_cache_free(revoke_table_cache, journal->j_revoke_table[0]);
-               return -ENOMEM;
+out:
+       return table;
+}
+
+static void journal_destroy_revoke_table(struct jbd_revoke_table_s *table)
+{
+       int i;
+       struct list_head *hash_list;
+
+       for (i = 0; i < table->hash_size; i++) {
+               hash_list = &table->hash_table[i];
+               J_ASSERT(list_empty(hash_list));
        }
 
-       journal->j_revoke = journal->j_revoke_table[1];
+       kfree(table->hash_table);
+       kmem_cache_free(revoke_table_cache, table);
+}
 
-       /* Check that the hash_size is a power of two */
+/* Initialise the revoke table for a given journal to a given size. */
+int journal_init_revoke(journal_t *journal, int hash_size)
+{
+       J_ASSERT(journal->j_revoke_table[0] == NULL);
        J_ASSERT(is_power_of_2(hash_size));
 
-       journal->j_revoke->hash_size = hash_size;
+       journal->j_revoke_table[0] = journal_init_revoke_table(hash_size);
+       if (!journal->j_revoke_table[0])
+               goto fail0;
 
-       journal->j_revoke->hash_shift = shift;
+       journal->j_revoke_table[1] = journal_init_revoke_table(hash_size);
+       if (!journal->j_revoke_table[1])
+               goto fail1;
 
-       journal->j_revoke->hash_table =
-               kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
-       if (!journal->j_revoke->hash_table) {
-               kfree(journal->j_revoke_table[0]->hash_table);
-               kmem_cache_free(revoke_table_cache, journal->j_revoke_table[0]);
-               kmem_cache_free(revoke_table_cache, journal->j_revoke_table[1]);
-               journal->j_revoke = NULL;
-               return -ENOMEM;
-       }
-
-       for (tmp = 0; tmp < hash_size; tmp++)
-               INIT_LIST_HEAD(&journal->j_revoke->hash_table[tmp]);
+       journal->j_revoke = journal->j_revoke_table[1];
 
        spin_lock_init(&journal->j_revoke_lock);
 
        return 0;
-}
 
-/* Destoy a journal's revoke table.  The table must already be empty! */
+fail1:
+       journal_destroy_revoke_table(journal->j_revoke_table[0]);
+fail0:
+       return -ENOMEM;
+}
 
+/* Destroy a journal's revoke table.  The table must already be empty! */
 void journal_destroy_revoke(journal_t *journal)
 {
-       struct jbd_revoke_table_s *table;
-       struct list_head *hash_list;
-       int i;
-
-       table = journal->j_revoke_table[0];
-       if (!table)
-               return;
-
-       for (i=0; i<table->hash_size; i++) {
-               hash_list = &table->hash_table[i];
-               J_ASSERT (list_empty(hash_list));
-       }
-
-       kfree(table->hash_table);
-       kmem_cache_free(revoke_table_cache, table);
-       journal->j_revoke = NULL;
-
-       table = journal->j_revoke_table[1];
-       if (!table)
-               return;
-
-       for (i=0; i<table->hash_size; i++) {
-               hash_list = &table->hash_table[i];
-               J_ASSERT (list_empty(hash_list));
-       }
-
-       kfree(table->hash_table);
-       kmem_cache_free(revoke_table_cache, table);
        journal->j_revoke = NULL;
+       if (journal->j_revoke_table[0])
+               journal_destroy_revoke_table(journal->j_revoke_table[0]);
+       if (journal->j_revoke_table[1])
+               journal_destroy_revoke_table(journal->j_revoke_table[1]);
 }
 
 
@@ -417,8 +422,6 @@ int journal_revoke(handle_t *handle, unsigned long blocknr,
  * the second time we would still have a pending revoke to cancel.  So,
  * do not trust the Revoked bit on buffers unless RevokeValid is also
  * set.
- *
- * The caller must have the journal locked.
  */
 int journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
 {
@@ -496,12 +499,9 @@ void journal_switch_revoke_table(journal_t *journal)
 /*
  * Write revoke records to the journal for all entries in the current
  * revoke hash, deleting the entries as we go.
- *
- * Called with the journal lock held.
  */
-
 void journal_write_revoke_records(journal_t *journal,
-                                 transaction_t *transaction)
+                                 transaction_t *transaction, int write_op)
 {
        struct journal_head *descriptor;
        struct jbd_revoke_record_s *record;
@@ -525,14 +525,14 @@ void journal_write_revoke_records(journal_t *journal,
                                hash_list->next;
                        write_one_revoke_record(journal, transaction,
                                                &descriptor, &offset,
-                                               record);
+                                               record, write_op);
                        count++;
                        list_del(&record->hash);
                        kmem_cache_free(revoke_record_cache, record);
                }
        }
        if (descriptor)
-               flush_descriptor(journal, descriptor, offset);
+               flush_descriptor(journal, descriptor, offset, write_op);
        jbd_debug(1, "Wrote %d revoke records\n", count);
 }
 
@@ -545,7 +545,8 @@ static void write_one_revoke_record(journal_t *journal,
                                    transaction_t *transaction,
                                    struct journal_head **descriptorp,
                                    int *offsetp,
-                                   struct jbd_revoke_record_s *record)
+                                   struct jbd_revoke_record_s *record,
+                                   int write_op)
 {
        struct journal_head *descriptor;
        int offset;
@@ -564,7 +565,7 @@ static void write_one_revoke_record(journal_t *journal,
        /* Make sure we have a descriptor with space left for the record */
        if (descriptor) {
                if (offset == journal->j_blocksize) {
-                       flush_descriptor(journal, descriptor, offset);
+                       flush_descriptor(journal, descriptor, offset, write_op);
                        descriptor = NULL;
                }
        }
@@ -601,7 +602,7 @@ static void write_one_revoke_record(journal_t *journal,
 
 static void flush_descriptor(journal_t *journal,
                             struct journal_head *descriptor,
-                            int offset)
+                            int offset, int write_op)
 {
        journal_revoke_header_t *header;
        struct buffer_head *bh = jh2bh(descriptor);
@@ -616,7 +617,7 @@ static void flush_descriptor(journal_t *journal,
        set_buffer_jwrite(bh);
        BUFFER_TRACE(bh, "write");
        set_buffer_dirty(bh);
-       ll_rw_block(SWRITE, 1, &bh);
+       ll_rw_block((write_op == WRITE) ? SWRITE : SWRITE_SYNC_PLUG, 1, &bh);
 }
 #endif