ext4: Use slab allocator for sub-page sized allocations

Now that the SLUB seems to be fixed so that it respects the requested
alignment, use kmem_cache_alloc() to allocator if the block size of
the buffer heads to be allocated is less than the page size.
Previously, we were using 16k page on a Power system for each buffer,
even when the file system was using 1k or 4k block size.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>

diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
index ac0d027..c03d4dc 100644 (file)
--- a/fs/jbd2/journal.c
+++ b/fs/jbd2/journal.c
@@ -39,6 +39,8 @@
 #include <linux/seq_file.h>
 #include <linux/math64.h>
 #include <linux/hash.h>
+#include <linux/log2.h>
+#include <linux/vmalloc.h>
 
 #define CREATE_TRACE_POINTS
 #include <trace/events/jbd2.h>
@@ -93,6 +95,7 @@ EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
 
 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
 static void __journal_abort_soft (journal_t *journal, int errno);
+static int jbd2_journal_create_slab(size_t slab_size);
 
 /*
  * Helper function used to manage commit timeouts
@@ -1248,6 +1251,13 @@ int jbd2_journal_load(journal_t *journal)
                }
        }
 
+       /*
+        * Create a slab for this blocksize
+        */
+       err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
+       if (err)
+               return err;
+
        /* Let the recovery code check whether it needs to recover any
         * data from the journal. */
        if (jbd2_journal_recover(journal))
@@ -1807,6 +1817,127 @@ size_t journal_tag_bytes(journal_t *journal)
 }
 
 /*
+ * JBD memory management
+ *
+ * These functions are used to allocate block-sized chunks of memory
+ * used for making copies of buffer_head data.  Very often it will be
+ * page-sized chunks of data, but sometimes it will be in
+ * sub-page-size chunks.  (For example, 16k pages on Power systems
+ * with a 4k block file system.)  For blocks smaller than a page, we
+ * use a SLAB allocator.  There are slab caches for each block size,
+ * which are allocated at mount time, if necessary, and we only free
+ * (all of) the slab caches when/if the jbd2 module is unloaded.  For
+ * this reason we don't need to a mutex to protect access to
+ * jbd2_slab[] allocating or releasing memory; only in
+ * jbd2_journal_create_slab().
+ */
+#define JBD2_MAX_SLABS 8
+static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
+static DECLARE_MUTEX(jbd2_slab_create_sem);
+
+static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
+       "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
+       "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
+};
+
+
+static void jbd2_journal_destroy_slabs(void)
+{
+       int i;
+
+       for (i = 0; i < JBD2_MAX_SLABS; i++) {
+               if (jbd2_slab[i])
+                       kmem_cache_destroy(jbd2_slab[i]);
+               jbd2_slab[i] = NULL;
+       }
+}
+
+static int jbd2_journal_create_slab(size_t size)
+{
+       int i = order_base_2(size) - 10;
+       size_t slab_size;
+
+       if (size == PAGE_SIZE)
+               return 0;
+
+       if (i >= JBD2_MAX_SLABS)
+               return -EINVAL;
+
+       if (unlikely(i < 0))
+               i = 0;
+       down(&jbd2_slab_create_sem);
+       if (jbd2_slab[i]) {
+               up(&jbd2_slab_create_sem);
+               return 0;       /* Already created */
+       }
+
+       slab_size = 1 << (i+10);
+       jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
+                                        slab_size, 0, NULL);
+       up(&jbd2_slab_create_sem);
+       if (!jbd2_slab[i]) {
+               printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
+               return -ENOMEM;
+       }
+       return 0;
+}
+
+static struct kmem_cache *get_slab(size_t size)
+{
+       int i = order_base_2(size) - 10;
+
+       BUG_ON(i >= JBD2_MAX_SLABS);
+       if (unlikely(i < 0))
+               i = 0;
+       BUG_ON(jbd2_slab[i] == 0);
+       return jbd2_slab[i];
+}
+
+void *jbd2_alloc(size_t size, gfp_t flags)
+{
+       void *ptr;
+
+       BUG_ON(size & (size-1)); /* Must be a power of 2 */
+
+       flags |= __GFP_REPEAT;
+       if (size == PAGE_SIZE)
+               ptr = (void *)__get_free_pages(flags, 0);
+       else if (size > PAGE_SIZE) {
+               int order = get_order(size);
+
+               if (order < 3)
+                       ptr = (void *)__get_free_pages(flags, order);
+               else
+                       ptr = vmalloc(size);
+       } else
+               ptr = kmem_cache_alloc(get_slab(size), flags);
+
+       /* Check alignment; SLUB has gotten this wrong in the past,
+        * and this can lead to user data corruption! */
+       BUG_ON(((unsigned long) ptr) & (size-1));
+
+       return ptr;
+}
+
+void jbd2_free(void *ptr, size_t size)
+{
+       if (size == PAGE_SIZE) {
+               free_pages((unsigned long)ptr, 0);
+               return;
+       }
+       if (size > PAGE_SIZE) {
+               int order = get_order(size);
+
+               if (order < 3)
+                       free_pages((unsigned long)ptr, order);
+               else
+                       vfree(ptr);
+               return;
+       }
+       kmem_cache_free(get_slab(size), ptr);
+};
+
+/*
  * Journal_head storage management
  */
 static struct kmem_cache *jbd2_journal_head_cache;
@@ -2204,6 +2335,7 @@ static void jbd2_journal_destroy_caches(void)
        jbd2_journal_destroy_revoke_caches();
        jbd2_journal_destroy_jbd2_journal_head_cache();
        jbd2_journal_destroy_handle_cache();
+       jbd2_journal_destroy_slabs();
 }
 
 static int __init journal_init(void)

diff --git a/include/linux/jbd2.h b/include/linux/jbd2.h
index 638ce45..8ada2a1 100644 (file)
--- a/include/linux/jbd2.h
+++ b/include/linux/jbd2.h
@@ -69,15 +69,8 @@ extern u8 jbd2_journal_enable_debug;
 #define jbd_debug(f, a...)     /**/
 #endif
 
-static inline void *jbd2_alloc(size_t size, gfp_t flags)
-{
-       return (void *)__get_free_pages(flags, get_order(size));
-}
-
-static inline void jbd2_free(void *ptr, size_t size)
-{
-       free_pages((unsigned long)ptr, get_order(size));
-};
+extern void *jbd2_alloc(size_t size, gfp_t flags);
+extern void jbd2_free(void *ptr, size_t size);
 
 #define JBD2_MIN_JOURNAL_BLOCKS 1024