Btrfs: Add ordered async work queues

Btrfs uses kernel threads to create async work queues for cpu intensive
operations such as checksumming and decompression.  These work well,
but they make it difficult to keep IO order intact.

A single writepages call from pdflush or fsync will turn into a number
of bios, and each bio is checksummed in parallel.  Once the checksum is
computed, the bio is sent down to the disk, and since we don't control
the order in which the parallel operations happen, they might go down to
the disk in almost any order.

The code deals with this somewhat by having deep work queues for a single
kernel thread, making it very likely that a single thread will process all
the bios for a single inode.

This patch introduces an explicitly ordered work queue.  As work structs
are placed into the queue they are put onto the tail of a list.  They have
three callbacks:

->func (cpu intensive processing here)
->ordered_func (order sensitive processing here)
->ordered_free (free the work struct, all processing is done)

The work struct has three callbacks.  The func callback does the cpu intensive
work, and when it completes the work struct is marked as done.

Every time a work struct completes, the list is checked to see if the head
is marked as done.  If so the ordered_func callback is used to do the
order sensitive processing and the ordered_free callback is used to do
any cleanup.  Then we loop back and check the head of the list again.

This patch also changes the checksumming code to use the ordered workqueues.
One a 4 drive array, it increases streaming writes from 280MB/s to 350MB/s.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

1 files changed, 64 insertions, 3 deletions

diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c
index d82efd722a48..e1e49715459e 100644
--- a/fs/btrfs/async-thread.c
+++ b/fs/btrfs/async-thread.c
@@ -23,6 +23,10 @@
 # include <linux/freezer.h>
 #include "async-thread.h"
 
+#define WORK_QUEUED_BIT 0
+#define WORK_DONE_BIT 1
+#define WORK_ORDER_DONE_BIT 2
+
 /*
  * container for the kthread task pointer and the list of pending work
  * One of these is allocated per thread.
@@ -88,6 +92,47 @@ static void check_busy_worker(struct btrfs_worker_thread *worker)
         }
 }
 
+static noinline int run_ordered_completions(struct btrfs_workers *workers,
+                                            struct btrfs_work *work)
+{
+        unsigned long flags;
+
+        if (!workers->ordered)
+                return 0;
+
+        set_bit(WORK_DONE_BIT, &work->flags);
+
+        spin_lock_irqsave(&workers->lock, flags);
+
+        while(!list_empty(&workers->order_list)) {
+                work = list_entry(workers->order_list.next,
+                                  struct btrfs_work, order_list);
+
+                if (!test_bit(WORK_DONE_BIT, &work->flags))
+                        break;
+
+                /* we are going to call the ordered done function, but
+                 * we leave the work item on the list as a barrier so
+                 * that later work items that are done don't have their
+                 * functions called before this one returns
+                 */
+                if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags))
+                        break;
+
+                spin_unlock_irqrestore(&workers->lock, flags);
+
+                work->ordered_func(work);
+
+                /* now take the lock again and call the freeing code */
+                spin_lock_irqsave(&workers->lock, flags);
+                list_del(&work->order_list);
+                work->ordered_free(work);
+        }
+
+        spin_unlock_irqrestore(&workers->lock, flags);
+        return 0;
+}
+
 /*
  * main loop for servicing work items
  */
@@ -102,7 +147,7 @@ static int worker_loop(void *arg)
                         cur = worker->pending.next;
                         work = list_entry(cur, struct btrfs_work, list);
                         list_del(&work->list);
-                        clear_bit(0, &work->flags);
+                        clear_bit(WORK_QUEUED_BIT, &work->flags);
 
                         work->worker = worker;
                         spin_unlock_irq(&worker->lock);
@@ -110,8 +155,15 @@ static int worker_loop(void *arg)
                         work->func(work);
 
                         atomic_dec(&worker->num_pending);
+                        /*
+                         * unless this is an ordered work queue,
+                         * 'work' was probably freed by func above.
+                         */
+                        run_ordered_completions(worker->workers, work);
+
                         spin_lock_irq(&worker->lock);
                         check_idle_worker(worker);
+
                 }
                 worker->working = 0;
                 if (freezing(current)) {
@@ -154,10 +206,12 @@ void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max)
         workers->num_workers = 0;
         INIT_LIST_HEAD(&workers->worker_list);
         INIT_LIST_HEAD(&workers->idle_list);
+        INIT_LIST_HEAD(&workers->order_list);
         spin_lock_init(&workers->lock);
         workers->max_workers = max;
         workers->idle_thresh = 32;
         workers->name = name;
+        workers->ordered = 0;
 }
 
 /*
@@ -296,7 +350,7 @@ int btrfs_requeue_work(struct btrfs_work *work)
         struct btrfs_worker_thread *worker = work->worker;
         unsigned long flags;
 
-        if (test_and_set_bit(0, &work->flags))
+        if (test_and_set_bit(WORK_QUEUED_BIT, &work->flags))
                 goto out;
 
         spin_lock_irqsave(&worker->lock, flags);
@@ -330,10 +384,17 @@ int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
         int wake = 0;
 
         /* don't requeue something already on a list */
-        if (test_and_set_bit(0, &work->flags))
+        if (test_and_set_bit(WORK_QUEUED_BIT, &work->flags))
                 goto out;
 
         worker = find_worker(workers);
+        if (workers->ordered) {
+                spin_lock_irqsave(&workers->lock, flags);
+                list_add_tail(&work->order_list, &workers->order_list);
+                spin_unlock_irqrestore(&workers->lock, flags);
+        } else {
+                INIT_LIST_HEAD(&work->order_list);
+        }
 
         spin_lock_irqsave(&worker->lock, flags);
         atomic_inc(&worker->num_pending);