diff mbox

[Bug,15018] ext4 backtraces out of nowhere

Message ID 201001201710.o0KHAXdu004099@demeter.kernel.org
State Superseded, archived
Headers show

Commit Message

bugzilla-daemon@bugzilla.kernel.org Jan. 20, 2010, 5:10 p.m. UTC 
http://bugzilla.kernel.org/show_bug.cgi?id=15018





--- Comment #3 from Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>  2010-01-20 17:10:31 ---
On Mon, 18 Jan 2010 23:06:08 GMT, bugzilla-daemon@bugzilla.kernel.org wrote:
> http://bugzilla.kernel.org/show_bug.cgi?id=15018
> 
> 
> Jan Kara <jack@suse.cz> changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |jack@suse.cz, tytso@mit.edu
> 
> 
> 
> 
> --- Comment #2 from Jan Kara <jack@suse.cz>  2010-01-18 23:06:07 ---
> Ah, inlining and such stuff made reading this stack trace quite difficult.
> Anyway, the problem is that we just should not call write_inode_now() in
> ext4_da_reserve_space() when we have transaction already started. And as
> Andreas correctly points out, the patch even introduces a deadlock because it
> can call write_inode_now with wait==1. Moreover we hold page lock when holding
> write_inode_now which introduces a plenty of nasty locking issues I believe.
> The bug has been introduced by commit 0637c6f4135f592f094207c7c21e7c0fc5557834.
> 
> Ted, I think that if we are failing to do the write because of ENOSPC, the only
> reasonable non-deadlocky thing to do is to restart the write completely - i.e.,
> bail out up to ext4_da_write_begin, stop the transaction, drop the page, *do
> some magic*, and start again...
> 

How about the below. I am yet to test the change

     unsigned from, to;
@@ -3091,6 +3079,21 @@ retry:

     if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
         goto retry;
+
+    if ((ret == -EDQUOT) && (quota_retries <= 3)) {
+        /*
+         * Since we do aggressive quota reservation may get no quota error
+         * even though we may be having quota. Forcing the inode write will
+         * free up the extra quota reserved. We do it only if we have
+         * meta data blocks reserved because we do aggressive reservation only
+         * for meta data blocks.
+         */
+        if (EXT4_I(inode)->i_reserved_meta_blocks) {
+            write_inode_now(inode, (quota_retries == 3));
+            quota_retries++;
+            goto retry;
+        }
+    }
 out:
     return ret;
 }
diff mbox

Patch

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index c818972..5af483a 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -1835,24 +1835,12 @@  repeat:
      * later. Real quota accounting is done at pages writeout
      * time.
      */
-    if (vfs_dq_reserve_block(inode, md_needed + 1)) {
-        /* 
-         * We tend to badly over-estimate the amount of
-         * metadata blocks which are needed, so if we have
-         * reserved any metadata blocks, try to force out the
-         * inode and see if we have any better luck.
-         */
-        if (md_reserved && retries++ <= 3)
-            goto retry;
+    if (vfs_dq_reserve_block(inode, md_needed + 1))
         return -EDQUOT;
-    }

     if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
         vfs_dq_release_reservation_block(inode, md_needed + 1);
         if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
-        retry:
-            if (md_reserved)
-                write_inode_now(inode, (retries == 3));
             yield();
             goto repeat;
         }
@@ -3032,7 +3020,7 @@  static int ext4_da_write_begin(struct file *file, struct
address_space *mapping,
                    loff_t pos, unsigned len, unsigned flags,
                    struct page **pagep, void **fsdata)
 {
-    int ret, retries = 0;
+    int ret, retries = 0, quota_retries = 0;
     struct page *page;
     pgoff_t index;