[PATCH] x86-64: Add __copy_from_user_nocache
[powerpc.git] / fs / buffer.c
index 517860f..f99c509 100644 (file)
@@ -35,6 +35,7 @@
 #include <linux/hash.h>
 #include <linux/suspend.h>
 #include <linux/buffer_head.h>
+#include <linux/task_io_accounting_ops.h>
 #include <linux/bio.h>
 #include <linux/notifier.h>
 #include <linux/cpu.h>
@@ -77,6 +78,7 @@ EXPORT_SYMBOL(__lock_buffer);
 
 void fastcall unlock_buffer(struct buffer_head *bh)
 {
+       smp_mb__before_clear_bit();
        clear_buffer_locked(bh);
        smp_mb__after_clear_bit();
        wake_up_bit(&bh->b_state, BH_Lock);
@@ -179,7 +181,7 @@ int fsync_bdev(struct block_device *bdev)
  * freeze_bdev  --  lock a filesystem and force it into a consistent state
  * @bdev:      blockdevice to lock
  *
- * This takes the block device bd_mount_mutex to make sure no new mounts
+ * This takes the block device bd_mount_sem to make sure no new mounts
  * happen on bdev until thaw_bdev() is called.
  * If a superblock is found on this device, we take the s_umount semaphore
  * on it to make sure nobody unmounts until the snapshot creation is done.
@@ -188,7 +190,7 @@ struct super_block *freeze_bdev(struct block_device *bdev)
 {
        struct super_block *sb;
 
-       mutex_lock(&bdev->bd_mount_mutex);
+       down(&bdev->bd_mount_sem);
        sb = get_super(bdev);
        if (sb && !(sb->s_flags & MS_RDONLY)) {
                sb->s_frozen = SB_FREEZE_WRITE;
@@ -230,7 +232,7 @@ void thaw_bdev(struct block_device *bdev, struct super_block *sb)
                drop_super(sb);
        }
 
-       mutex_unlock(&bdev->bd_mount_mutex);
+       up(&bdev->bd_mount_sem);
 }
 EXPORT_SYMBOL(thaw_bdev);
 
@@ -344,7 +346,7 @@ void invalidate_bdev(struct block_device *bdev, int destroy_dirty_buffers)
         * We really want to use invalidate_inode_pages2() for
         * that, but not until that's cleaned up.
         */
-       invalidate_inode_pages(mapping);
+       invalidate_mapping_pages(mapping, 0, -1);
 }
 
 /*
@@ -724,20 +726,21 @@ int __set_page_dirty_buffers(struct page *page)
        }
        spin_unlock(&mapping->private_lock);
 
-       if (!TestSetPageDirty(page)) {
-               write_lock_irq(&mapping->tree_lock);
-               if (page->mapping) {    /* Race with truncate? */
-                       if (mapping_cap_account_dirty(mapping))
-                               __inc_zone_page_state(page, NR_FILE_DIRTY);
-                       radix_tree_tag_set(&mapping->page_tree,
-                                               page_index(page),
-                                               PAGECACHE_TAG_DIRTY);
+       if (TestSetPageDirty(page))
+               return 0;
+
+       write_lock_irq(&mapping->tree_lock);
+       if (page->mapping) {    /* Race with truncate? */
+               if (mapping_cap_account_dirty(mapping)) {
+                       __inc_zone_page_state(page, NR_FILE_DIRTY);
+                       task_io_account_write(PAGE_CACHE_SIZE);
                }
-               write_unlock_irq(&mapping->tree_lock);
-               __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
-               return 1;
+               radix_tree_tag_set(&mapping->page_tree,
+                               page_index(page), PAGECACHE_TAG_DIRTY);
        }
-       return 0;
+       write_unlock_irq(&mapping->tree_lock);
+       __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
+       return 1;
 }
 EXPORT_SYMBOL(__set_page_dirty_buffers);
 
@@ -1280,11 +1283,11 @@ static void bh_lru_install(struct buffer_head *bh)
  * Look up the bh in this cpu's LRU.  If it's there, move it to the head.
  */
 static struct buffer_head *
-lookup_bh_lru(struct block_device *bdev, sector_t block, int size)
+lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
 {
        struct buffer_head *ret = NULL;
        struct bh_lru *lru;
-       int i;
+       unsigned int i;
 
        check_irqs_on();
        bh_lru_lock();
@@ -1316,7 +1319,7 @@ lookup_bh_lru(struct block_device *bdev, sector_t block, int size)
  * NULL
  */
 struct buffer_head *
-__find_get_block(struct block_device *bdev, sector_t block, int size)
+__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
 {
        struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
 
@@ -1344,7 +1347,7 @@ EXPORT_SYMBOL(__find_get_block);
  * attempt is failing.  FIXME, perhaps?
  */
 struct buffer_head *
-__getblk(struct block_device *bdev, sector_t block, int size)
+__getblk(struct block_device *bdev, sector_t block, unsigned size)
 {
        struct buffer_head *bh = __find_get_block(bdev, block, size);
 
@@ -1358,7 +1361,7 @@ EXPORT_SYMBOL(__getblk);
 /*
  * Do async read-ahead on a buffer..
  */
-void __breadahead(struct block_device *bdev, sector_t block, int size)
+void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
 {
        struct buffer_head *bh = __getblk(bdev, block, size);
        if (likely(bh)) {
@@ -1378,7 +1381,7 @@ EXPORT_SYMBOL(__breadahead);
  *  It returns NULL if the block was unreadable.
  */
 struct buffer_head *
-__bread(struct block_device *bdev, sector_t block, int size)
+__bread(struct block_device *bdev, sector_t block, unsigned size)
 {
        struct buffer_head *bh = __getblk(bdev, block, size);
 
@@ -1437,6 +1440,7 @@ static void discard_buffer(struct buffer_head * bh)
        clear_buffer_req(bh);
        clear_buffer_new(bh);
        clear_buffer_delay(bh);
+       clear_buffer_unwritten(bh);
        unlock_buffer(bh);
 }
 
@@ -1820,6 +1824,7 @@ static int __block_prepare_write(struct inode *inode, struct page *page,
                        continue; 
                }
                if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
+                   !buffer_unwritten(bh) &&
                     (block_start < from || block_end > to)) {
                        ll_rw_block(READ, 1, &bh);
                        *wait_bh++=bh;
@@ -2541,7 +2546,7 @@ int block_truncate_page(struct address_space *mapping,
        if (PageUptodate(page))
                set_buffer_uptodate(bh);
 
-       if (!buffer_uptodate(bh) && !buffer_delay(bh)) {
+       if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
                err = -EIO;
                ll_rw_block(READ, 1, &bh);
                wait_on_buffer(bh);
@@ -2842,18 +2847,24 @@ int try_to_free_buffers(struct page *page)
 
        spin_lock(&mapping->private_lock);
        ret = drop_buffers(page, &buffers_to_free);
+
+       /*
+        * If the filesystem writes its buffers by hand (eg ext3)
+        * then we can have clean buffers against a dirty page.  We
+        * clean the page here; otherwise the VM will never notice
+        * that the filesystem did any IO at all.
+        *
+        * Also, during truncate, discard_buffer will have marked all
+        * the page's buffers clean.  We discover that here and clean
+        * the page also.
+        *
+        * private_lock must be held over this entire operation in order
+        * to synchronise against __set_page_dirty_buffers and prevent the
+        * dirty bit from being lost.
+        */
+       if (ret)
+               cancel_dirty_page(page, PAGE_CACHE_SIZE);
        spin_unlock(&mapping->private_lock);
-       if (ret) {
-               /*
-                * If the filesystem writes its buffers by hand (eg ext3)
-                * then we can have clean buffers against a dirty page.  We
-                * clean the page here; otherwise later reattachment of buffers
-                * could encounter a non-uptodate page, which is unresolvable.
-                * This only applies in the rare case where try_to_free_buffers
-                * succeeds but the page is not freed.
-                */
-               clear_page_dirty(page);
-       }
 out:
        if (buffers_to_free) {
                struct buffer_head *bh = buffers_to_free;