fuse: postpone end_page_writeback() in fuse_writepage_locked()
[linux-3.10.git] / fs / fs-writeback.c
index fd255c0..3be5718 100644 (file)
@@ -22,7 +22,6 @@
 #include <linux/mm.h>
 #include <linux/pagemap.h>
 #include <linux/kthread.h>
-#include <linux/freezer.h>
 #include <linux/writeback.h>
 #include <linux/blkdev.h>
 #include <linux/backing-dev.h>
@@ -63,6 +62,7 @@ int writeback_in_progress(struct backing_dev_info *bdi)
 {
        return test_bit(BDI_writeback_running, &bdi->state);
 }
+EXPORT_SYMBOL(writeback_in_progress);
 
 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
 {
@@ -87,20 +87,6 @@ static inline struct inode *wb_inode(struct list_head *head)
 #define CREATE_TRACE_POINTS
 #include <trace/events/writeback.h>
 
-/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
-static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
-{
-       if (bdi->wb.task) {
-               wake_up_process(bdi->wb.task);
-       } else {
-               /*
-                * The bdi thread isn't there, wake up the forker thread which
-                * will create and run it.
-                */
-               wake_up_process(default_backing_dev_info.wb.task);
-       }
-}
-
 static void bdi_queue_work(struct backing_dev_info *bdi,
                           struct wb_writeback_work *work)
 {
@@ -108,10 +94,9 @@ static void bdi_queue_work(struct backing_dev_info *bdi,
 
        spin_lock_bh(&bdi->wb_lock);
        list_add_tail(&work->list, &bdi->work_list);
-       if (!bdi->wb.task)
-               trace_writeback_nothread(bdi, work);
-       bdi_wakeup_flusher(bdi);
        spin_unlock_bh(&bdi->wb_lock);
+
+       mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 }
 
 static void
@@ -126,10 +111,8 @@ __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
         */
        work = kzalloc(sizeof(*work), GFP_ATOMIC);
        if (!work) {
-               if (bdi->wb.task) {
-                       trace_writeback_nowork(bdi);
-                       wake_up_process(bdi->wb.task);
-               }
+               trace_writeback_nowork(bdi);
+               mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
                return;
        }
 
@@ -176,9 +159,7 @@ void bdi_start_background_writeback(struct backing_dev_info *bdi)
         * writeback as soon as there is no other work to do.
         */
        trace_writeback_wake_background(bdi);
-       spin_lock_bh(&bdi->wb_lock);
-       bdi_wakeup_flusher(bdi);
-       spin_unlock_bh(&bdi->wb_lock);
+       mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 }
 
 /*
@@ -227,6 +208,8 @@ static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
 static void inode_sync_complete(struct inode *inode)
 {
        inode->i_state &= ~I_SYNC;
+       /* If inode is clean an unused, put it into LRU now... */
+       inode_add_lru(inode);
        /* Waiters must see I_SYNC cleared before being woken up */
        smp_mb();
        wake_up_bit(&inode->i_state, __I_SYNC);
@@ -315,8 +298,14 @@ static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
 
 static int write_inode(struct inode *inode, struct writeback_control *wbc)
 {
-       if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
-               return inode->i_sb->s_op->write_inode(inode, wbc);
+       int ret;
+
+       if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
+               trace_writeback_write_inode_start(inode, wbc);
+               ret = inode->i_sb->s_op->write_inode(inode, wbc);
+               trace_writeback_write_inode(inode, wbc);
+               return ret;
+       }
        return 0;
 }
 
@@ -438,8 +427,7 @@ static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
  * setting I_SYNC flag and calling inode_sync_complete() to clear it.
  */
 static int
-__writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
-                        struct writeback_control *wbc)
+__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
 {
        struct address_space *mapping = inode->i_mapping;
        long nr_to_write = wbc->nr_to_write;
@@ -448,6 +436,8 @@ __writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
 
        WARN_ON(!(inode->i_state & I_SYNC));
 
+       trace_writeback_single_inode_start(inode, wbc, nr_to_write);
+
        ret = do_writepages(mapping, wbc);
 
        /*
@@ -526,7 +516,7 @@ writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
        inode->i_state |= I_SYNC;
        spin_unlock(&inode->i_lock);
 
-       ret = __writeback_single_inode(inode, wb, wbc);
+       ret = __writeback_single_inode(inode, wbc);
 
        spin_lock(&wb->list_lock);
        spin_lock(&inode->i_lock);
@@ -577,10 +567,6 @@ static long writeback_chunk_size(struct backing_dev_info *bdi,
 /*
  * Write a portion of b_io inodes which belong to @sb.
  *
- * If @only_this_sb is true, then find and write all such
- * inodes. Otherwise write only ones which go sequentially
- * in reverse order.
- *
  * Return the number of pages and/or inodes written.
  */
 static long writeback_sb_inodes(struct super_block *sb,
@@ -673,7 +659,7 @@ static long writeback_sb_inodes(struct super_block *sb,
                 * We use I_SYNC to pin the inode in memory. While it is set
                 * evict_inode() will wait so the inode cannot be freed.
                 */
-               __writeback_single_inode(inode, wb, &wbc);
+               __writeback_single_inode(inode, &wbc);
 
                work->nr_pages -= write_chunk - wbc.nr_to_write;
                wrote += write_chunk - wbc.nr_to_write;
@@ -1014,66 +1000,49 @@ long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
 
 /*
  * Handle writeback of dirty data for the device backed by this bdi. Also
- * wakes up periodically and does kupdated style flushing.
+ * reschedules periodically and does kupdated style flushing.
  */
-int bdi_writeback_thread(void *data)
+void bdi_writeback_workfn(struct work_struct *work)
 {
-       struct bdi_writeback *wb = data;
+       struct bdi_writeback *wb = container_of(to_delayed_work(work),
+                                               struct bdi_writeback, dwork);
        struct backing_dev_info *bdi = wb->bdi;
        long pages_written;
 
+       set_worker_desc("flush-%s", dev_name(bdi->dev));
        current->flags |= PF_SWAPWRITE;
-       set_freezable();
-       wb->last_active = jiffies;
 
-       /*
-        * Our parent may run at a different priority, just set us to normal
-        */
-       set_user_nice(current, 0);
-
-       trace_writeback_thread_start(bdi);
-
-       while (!kthread_freezable_should_stop(NULL)) {
+       if (likely(!current_is_workqueue_rescuer() ||
+                  list_empty(&bdi->bdi_list))) {
                /*
-                * Remove own delayed wake-up timer, since we are already awake
-                * and we'll take care of the preriodic write-back.
+                * The normal path.  Keep writing back @bdi until its
+                * work_list is empty.  Note that this path is also taken
+                * if @bdi is shutting down even when we're running off the
+                * rescuer as work_list needs to be drained.
                 */
-               del_timer(&wb->wakeup_timer);
-
-               pages_written = wb_do_writeback(wb, 0);
-
+               do {
+                       pages_written = wb_do_writeback(wb, 0);
+                       trace_writeback_pages_written(pages_written);
+               } while (!list_empty(&bdi->work_list));
+       } else {
+               /*
+                * bdi_wq can't get enough workers and we're running off
+                * the emergency worker.  Don't hog it.  Hopefully, 1024 is
+                * enough for efficient IO.
+                */
+               pages_written = writeback_inodes_wb(&bdi->wb, 1024,
+                                                   WB_REASON_FORKER_THREAD);
                trace_writeback_pages_written(pages_written);
-
-               if (pages_written)
-                       wb->last_active = jiffies;
-
-               set_current_state(TASK_INTERRUPTIBLE);
-               if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
-                       __set_current_state(TASK_RUNNING);
-                       continue;
-               }
-
-               if (wb_has_dirty_io(wb) && dirty_writeback_interval)
-                       schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
-               else {
-                       /*
-                        * We have nothing to do, so can go sleep without any
-                        * timeout and save power. When a work is queued or
-                        * something is made dirty - we will be woken up.
-                        */
-                       schedule();
-               }
        }
 
-       /* Flush any work that raced with us exiting */
-       if (!list_empty(&bdi->work_list))
-               wb_do_writeback(wb, 1);
+       if (!list_empty(&bdi->work_list) ||
+           (wb_has_dirty_io(wb) && dirty_writeback_interval))
+               queue_delayed_work(bdi_wq, &wb->dwork,
+                       msecs_to_jiffies(dirty_writeback_interval * 10));
 
-       trace_writeback_thread_stop(bdi);
-       return 0;
+       current->flags &= ~PF_SWAPWRITE;
 }
 
-
 /*
  * Start writeback of `nr_pages' pages.  If `nr_pages' is zero, write back
  * the whole world.
@@ -1152,8 +1121,12 @@ void __mark_inode_dirty(struct inode *inode, int flags)
         * dirty the inode itself
         */
        if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
+               trace_writeback_dirty_inode_start(inode, flags);
+
                if (sb->s_op->dirty_inode)
                        sb->s_op->dirty_inode(inode, flags);
+
+               trace_writeback_dirty_inode(inode, flags);
        }
 
        /*
@@ -1334,47 +1307,43 @@ void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
 EXPORT_SYMBOL(writeback_inodes_sb);
 
 /**
- * writeback_inodes_sb_if_idle -       start writeback if none underway
+ * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
  * @sb: the superblock
- * @reason: reason why some writeback work was initiated
+ * @nr: the number of pages to write
+ * @reason: the reason of writeback
  *
- * Invoke writeback_inodes_sb if no writeback is currently underway.
+ * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
  * Returns 1 if writeback was started, 0 if not.
  */
-int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
+int try_to_writeback_inodes_sb_nr(struct super_block *sb,
+                                 unsigned long nr,
+                                 enum wb_reason reason)
 {
-       if (!writeback_in_progress(sb->s_bdi)) {
-               down_read(&sb->s_umount);
-               writeback_inodes_sb(sb, reason);
-               up_read(&sb->s_umount);
+       if (writeback_in_progress(sb->s_bdi))
                return 1;
-       } else
+
+       if (!down_read_trylock(&sb->s_umount))
                return 0;
+
+       writeback_inodes_sb_nr(sb, nr, reason);
+       up_read(&sb->s_umount);
+       return 1;
 }
-EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
+EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
 
 /**
- * writeback_inodes_sb_nr_if_idle      -       start writeback if none underway
+ * try_to_writeback_inodes_sb - try to start writeback if none underway
  * @sb: the superblock
- * @nr: the number of pages to write
  * @reason: reason why some writeback work was initiated
  *
- * Invoke writeback_inodes_sb if no writeback is currently underway.
+ * Implement by try_to_writeback_inodes_sb_nr()
  * Returns 1 if writeback was started, 0 if not.
  */
-int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
-                                  unsigned long nr,
-                                  enum wb_reason reason)
+int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
 {
-       if (!writeback_in_progress(sb->s_bdi)) {
-               down_read(&sb->s_umount);
-               writeback_inodes_sb_nr(sb, nr, reason);
-               up_read(&sb->s_umount);
-               return 1;
-       } else
-               return 0;
+       return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
 }
-EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
+EXPORT_SYMBOL(try_to_writeback_inodes_sb);
 
 /**
  * sync_inodes_sb      -       sync sb inode pages