[PATCH] md: improve locking on 'safemode' and move superblock writes

When md marks the superblock dirty before a write, it calls
generic_make_request (to write the superblock) from within
generic_make_request (to write the first dirty block), which could cause
problems later.

With this patch, the superblock write is always done by the helper thread, and
write request are delayed until that write completes.

Also, the locking around marking the array dirty and writing the superblock is
improved to avoid possible races.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

1 files changed, 59 insertions, 14 deletions

diff --git a/drivers/md/md.c b/drivers/md/md.c
index c842e34d850e..177d2a7d7cea 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -218,6 +218,8 @@ static mddev_t * mddev_find(dev_t unit)
         INIT_LIST_HEAD(&new->all_mddevs);
         init_timer(&new->safemode_timer);
         atomic_set(&new->active, 1);
+        bio_list_init(&new->write_list);
+        spin_lock_init(&new->write_lock);
 
         new->queue = blk_alloc_queue(GFP_KERNEL);
         if (!new->queue) {
@@ -1251,9 +1253,11 @@ static void md_update_sb(mddev_t * mddev)
         int err, count = 100;
         struct list_head *tmp;
         mdk_rdev_t *rdev;
+        int sync_req;
 
-        mddev->sb_dirty = 0;
 repeat:
+        spin_lock(&mddev->write_lock);
+        sync_req = mddev->in_sync;
         mddev->utime = get_seconds();
         mddev->events ++;
 
@@ -1272,8 +1276,12 @@ repeat:
          * do not write anything to disk if using
          * nonpersistent superblocks
          */
-        if (!mddev->persistent)
+        if (!mddev->persistent) {
+                mddev->sb_dirty = 0;
+                spin_unlock(&mddev->write_lock);
                 return;
+        }
+        spin_unlock(&mddev->write_lock);
 
         dprintk(KERN_INFO 
                 "md: updating %s RAID superblock on device (in sync %d)\n",
@@ -1304,6 +1312,15 @@ repeat:
                 printk(KERN_ERR \
                         "md: excessive errors occurred during superblock update, exiting\n");
         }
+        spin_lock(&mddev->write_lock);
+        if (mddev->in_sync != sync_req) {
+                /* have to write it out again */
+                spin_unlock(&mddev->write_lock);
+                goto repeat;
+        }
+        mddev->sb_dirty = 0;
+        spin_unlock(&mddev->write_lock);
+
 }
 
 /*
@@ -3178,19 +3195,31 @@ void md_done_sync(mddev_t *mddev, int blocks, int ok)
 }
 
 
-void md_write_start(mddev_t *mddev)
+/* md_write_start(mddev, bi)
+ * If we need to update some array metadata (e.g. 'active' flag
+ * in superblock) before writing, queue bi for later writing
+ * and return 0, else return 1 and it will be written now
+ */
+int md_write_start(mddev_t *mddev, struct bio *bi)
 {
-        if (!atomic_read(&mddev->writes_pending)) {
-                mddev_lock_uninterruptible(mddev);
-                if (mddev->in_sync) {
-                        mddev->in_sync = 0;
-                        del_timer(&mddev->safemode_timer);
-                        md_update_sb(mddev);
-                }
-                atomic_inc(&mddev->writes_pending);
-                mddev_unlock(mddev);
-        } else
-                atomic_inc(&mddev->writes_pending);
+        if (bio_data_dir(bi) != WRITE)
+                return 1;
+
+        atomic_inc(&mddev->writes_pending);
+        spin_lock(&mddev->write_lock);
+        if (mddev->in_sync == 0 && mddev->sb_dirty == 0) {
+                spin_unlock(&mddev->write_lock);
+                return 1;
+        }
+        bio_list_add(&mddev->write_list, bi);
+
+        if (mddev->in_sync) {
+                mddev->in_sync = 0;
+                mddev->sb_dirty = 1;
+        }
+        spin_unlock(&mddev->write_lock);
+        md_wakeup_thread(mddev->thread);
+        return 0;
 }
 
 void md_write_end(mddev_t *mddev)
@@ -3472,6 +3501,7 @@ void md_check_recovery(mddev_t *mddev)
                 mddev->sb_dirty ||
                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
+                mddev->write_list.head ||
                 (mddev->safemode == 1) ||
                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
@@ -3480,7 +3510,9 @@ void md_check_recovery(mddev_t *mddev)
 
         if (mddev_trylock(mddev)==0) {
                 int spares =0;
+                struct bio *blist;
 
+                spin_lock(&mddev->write_lock);
                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
                         mddev->in_sync = 1;
@@ -3488,9 +3520,22 @@ void md_check_recovery(mddev_t *mddev)
                 }
                 if (mddev->safemode == 1)
                         mddev->safemode = 0;
+                blist = bio_list_get(&mddev->write_list);
+                spin_unlock(&mddev->write_lock);
 
                 if (mddev->sb_dirty)
                         md_update_sb(mddev);
+
+                while (blist) {
+                        struct bio *b = blist;
+                        blist = blist->bi_next;
+                        b->bi_next = NULL;
+                        generic_make_request(b);
+                        /* we already counted this, so need to un-count */
+                        md_write_end(mddev);
+                }
+
+
                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
                         /* resync/recovery still happening */