Pull ia64-clocksource into release branch
[pandora-kernel.git] / mm / page-writeback.c
index 029dfad..63512a9 100644 (file)
@@ -476,15 +476,13 @@ static void wb_kupdate(unsigned long arg)
  * sysctl handler for /proc/sys/vm/dirty_writeback_centisecs
  */
 int dirty_writeback_centisecs_handler(ctl_table *table, int write,
-               struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
+       struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
 {
        proc_dointvec_userhz_jiffies(table, write, file, buffer, length, ppos);
-       if (dirty_writeback_interval) {
-               mod_timer(&wb_timer,
-                       jiffies + dirty_writeback_interval);
-               } else {
+       if (dirty_writeback_interval)
+               mod_timer(&wb_timer, jiffies + dirty_writeback_interval);
+       else
                del_timer(&wb_timer);
-       }
        return 0;
 }
 
@@ -588,31 +586,27 @@ void __init page_writeback_init(void)
 }
 
 /**
- * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them.
+ * write_cache_pages - walk the list of dirty pages of the given address space and write all of them.
  * @mapping: address space structure to write
  * @wbc: subtract the number of written pages from *@wbc->nr_to_write
+ * @writepage: function called for each page
+ * @data: data passed to writepage function
  *
- * This is a library function, which implements the writepages()
- * address_space_operation.
- *
- * If a page is already under I/O, generic_writepages() skips it, even
+ * If a page is already under I/O, write_cache_pages() skips it, even
  * if it's dirty.  This is desirable behaviour for memory-cleaning writeback,
  * but it is INCORRECT for data-integrity system calls such as fsync().  fsync()
  * and msync() need to guarantee that all the data which was dirty at the time
  * the call was made get new I/O started against them.  If wbc->sync_mode is
  * WB_SYNC_ALL then we were called for data integrity and we must wait for
  * existing IO to complete.
- *
- * Derived from mpage_writepages() - if you fix this you should check that
- * also!
  */
-int generic_writepages(struct address_space *mapping,
-                      struct writeback_control *wbc)
+int write_cache_pages(struct address_space *mapping,
+                     struct writeback_control *wbc, writepage_t writepage,
+                     void *data)
 {
        struct backing_dev_info *bdi = mapping->backing_dev_info;
        int ret = 0;
        int done = 0;
-       int (*writepage)(struct page *page, struct writeback_control *wbc);
        struct pagevec pvec;
        int nr_pages;
        pgoff_t index;
@@ -625,12 +619,6 @@ int generic_writepages(struct address_space *mapping,
                return 0;
        }
 
-       writepage = mapping->a_ops->writepage;
-
-       /* deal with chardevs and other special file */
-       if (!writepage)
-               return 0;
-
        pagevec_init(&pvec, 0);
        if (wbc->range_cyclic) {
                index = mapping->writeback_index; /* Start from prev offset */
@@ -682,13 +670,7 @@ retry:
                                continue;
                        }
 
-                       ret = (*writepage)(page, wbc);
-                       if (ret) {
-                               if (ret == -ENOSPC)
-                                       set_bit(AS_ENOSPC, &mapping->flags);
-                               else
-                                       set_bit(AS_EIO, &mapping->flags);
-                       }
+                       ret = (*writepage)(page, wbc, data);
 
                        if (unlikely(ret == AOP_WRITEPAGE_ACTIVATE))
                                unlock_page(page);
@@ -715,6 +697,38 @@ retry:
                mapping->writeback_index = index;
        return ret;
 }
+EXPORT_SYMBOL(write_cache_pages);
+
+/*
+ * Function used by generic_writepages to call the real writepage
+ * function and set the mapping flags on error
+ */
+static int __writepage(struct page *page, struct writeback_control *wbc,
+                      void *data)
+{
+       struct address_space *mapping = data;
+       int ret = mapping->a_ops->writepage(page, wbc);
+       mapping_set_error(mapping, ret);
+       return ret;
+}
+
+/**
+ * generic_writepages - walk the list of dirty pages of the given address space and writepage() all of them.
+ * @mapping: address space structure to write
+ * @wbc: subtract the number of written pages from *@wbc->nr_to_write
+ *
+ * This is a library function, which implements the writepages()
+ * address_space_operation.
+ */
+int generic_writepages(struct address_space *mapping,
+                      struct writeback_control *wbc)
+{
+       /* deal with chardevs and other special file */
+       if (!mapping->a_ops->writepage)
+               return 0;
+
+       return write_cache_pages(mapping, wbc, __writepage, mapping);
+}
 
 EXPORT_SYMBOL(generic_writepages);
 
@@ -810,6 +824,7 @@ int __set_page_dirty_nobuffers(struct page *page)
                mapping2 = page_mapping(page);
                if (mapping2) { /* Race with truncate? */
                        BUG_ON(mapping2 != mapping);
+                       WARN_ON_ONCE(!PagePrivate(page) && !PageUptodate(page));
                        if (mapping_cap_account_dirty(mapping)) {
                                __inc_zone_page_state(page, NR_FILE_DIRTY);
                                task_io_account_write(PAGE_CACHE_SIZE);
@@ -903,6 +918,9 @@ int clear_page_dirty_for_io(struct page *page)
 {
        struct address_space *mapping = page_mapping(page);
 
+       BUG_ON(!PageLocked(page));
+
+       ClearPageReclaim(page);
        if (mapping && mapping_cap_account_dirty(mapping)) {
                /*
                 * Yes, Virginia, this is indeed insane.
@@ -928,14 +946,19 @@ int clear_page_dirty_for_io(struct page *page)
                 * We basically use the page "master dirty bit"
                 * as a serialization point for all the different
                 * threads doing their things.
-                *
-                * FIXME! We still have a race here: if somebody
-                * adds the page back to the page tables in
-                * between the "page_mkclean()" and the "TestClearPageDirty()",
-                * we might have it mapped without the dirty bit set.
                 */
                if (page_mkclean(page))
                        set_page_dirty(page);
+               /*
+                * We carefully synchronise fault handlers against
+                * installing a dirty pte and marking the page dirty
+                * at this point. We do this by having them hold the
+                * page lock at some point after installing their
+                * pte, but before marking the page dirty.
+                * Pages are always locked coming in here, so we get
+                * the desired exclusion. See mm/memory.c:do_wp_page()
+                * for more comments.
+                */
                if (TestClearPageDirty(page)) {
                        dec_zone_page_state(page, NR_FILE_DIRTY);
                        return 1;
@@ -964,6 +987,8 @@ int test_clear_page_writeback(struct page *page)
        } else {
                ret = TestClearPageWriteback(page);
        }
+       if (ret)
+               dec_zone_page_state(page, NR_WRITEBACK);
        return ret;
 }
 
@@ -989,6 +1014,8 @@ int test_set_page_writeback(struct page *page)
        } else {
                ret = TestSetPageWriteback(page);
        }
+       if (!ret)
+               inc_zone_page_state(page, NR_WRITEBACK);
        return ret;
 
 }