]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
[XFS] Remove xfs_iext_irec_compact_full()
authorLachlan McIlroy <lachlan@sgi.com>
Fri, 26 Sep 2008 02:17:57 +0000 (12:17 +1000)
committerLachlan McIlroy <lachlan@redback.melbourne.sgi.com>
Fri, 26 Sep 2008 02:17:57 +0000 (12:17 +1000)
Yet another bug was found in xfs_iext_irec_compact_full() and while the
source of the bug was found it wasn't an easy task to track it down
because the conditions are very difficult to reproduce.

A HUGE thank-you goes to Russell Cattelan and Eric Sandeen for their
significant effort in tracking down the source of this corruption.

xfs_iext_irec_compact_full() and xfs_iext_irec_compact_pages() are almost
identical - they both compact indirect extent lists by moving extents from
subsequent buffers into earlier ones. xfs_iext_irec_compact_pages() only
moves extents if all of the extents in the next buffer will fit into the
empty space in the buffer before it. xfs_iext_irec_compact_full() will go
a step further and move part of the next buffer if all the extents wont
fit. It will then shift the remaining extents in the next buffer up to the
start of the buffer. The bug here was that we did not update er_extoff and
this caused extent list corruption.

It does not appear that this extra functionality gains us much. Calling
xfs_iext_irec_compact_pages() instead will do a good enough job at
compacting the indirect list and will be quicker too.

For the case in xfs_iext_indirect_to_direct() the total number of extents
in the indirect list will fit into one buffer so we will never need the
extra functionality of xfs_iext_irec_compact_full() there.

Also xfs_iext_irec_compact_pages() doesn't need to do a memmove() (the
buffers will never overlap) so we don't want the performance hit that can
incur.

SGI-PV: 987159

SGI-Modid: xfs-linux-melb:xfs-kern:32166a

Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Eric Sandeen <sandeen@sandeen.net>
fs/xfs/xfs_inode.c

index 419cfc2eacb3ae522b00e7ce0750252947d8ab8b..dbd9cef852ece3d6d8b3b5b930626000c88bed06 100644 (file)
@@ -4118,7 +4118,7 @@ xfs_iext_indirect_to_direct(
        ASSERT(nextents <= XFS_LINEAR_EXTS);
        size = nextents * sizeof(xfs_bmbt_rec_t);
 
-       xfs_iext_irec_compact_full(ifp);
+       xfs_iext_irec_compact_pages(ifp);
        ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
 
        ep = ifp->if_u1.if_ext_irec->er_extbuf;
@@ -4449,8 +4449,7 @@ xfs_iext_irec_remove(
  * compaction policy is as follows:
  *
  *    Full Compaction: Extents fit into a single page (or inline buffer)
- *    Full Compaction: Extents occupy less than 10% of allocated space
- * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
+ * Partial Compaction: Extents occupy less than 50% of allocated space
  *      No Compaction: Extents occupy at least 50% of allocated space
  */
 void
@@ -4471,8 +4470,6 @@ xfs_iext_irec_compact(
                xfs_iext_direct_to_inline(ifp, nextents);
        } else if (nextents <= XFS_LINEAR_EXTS) {
                xfs_iext_indirect_to_direct(ifp);
-       } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
-               xfs_iext_irec_compact_full(ifp);
        } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
                xfs_iext_irec_compact_pages(ifp);
        }
@@ -4496,7 +4493,7 @@ xfs_iext_irec_compact_pages(
                erp_next = erp + 1;
                if (erp_next->er_extcount <=
                    (XFS_LINEAR_EXTS - erp->er_extcount)) {
-                       memmove(&erp->er_extbuf[erp->er_extcount],
+                       memcpy(&erp->er_extbuf[erp->er_extcount],
                                erp_next->er_extbuf, erp_next->er_extcount *
                                sizeof(xfs_bmbt_rec_t));
                        erp->er_extcount += erp_next->er_extcount;
@@ -4515,92 +4512,6 @@ xfs_iext_irec_compact_pages(
        }
 }
 
-/*
- * Fully compact the extent records managed by the indirection array.
- */
-void
-xfs_iext_irec_compact_full(
-       xfs_ifork_t     *ifp)                   /* inode fork pointer */
-{
-       xfs_bmbt_rec_host_t *ep, *ep_next;      /* extent record pointers */
-       xfs_ext_irec_t  *erp, *erp_next;        /* extent irec pointers */
-       int             erp_idx = 0;            /* extent irec index */
-       int             ext_avail;              /* empty entries in ex list */
-       int             ext_diff;               /* number of exts to add */
-       int             nlists;                 /* number of irec's (ex lists) */
-
-       ASSERT(ifp->if_flags & XFS_IFEXTIREC);
-
-       nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
-       erp = ifp->if_u1.if_ext_irec;
-       ep = &erp->er_extbuf[erp->er_extcount];
-       erp_next = erp + 1;
-       ep_next = erp_next->er_extbuf;
-
-       while (erp_idx < nlists - 1) {
-               /*
-                * Check how many extent records are available in this irec.
-                * If there is none skip the whole exercise.
-                */
-               ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
-               if (ext_avail) {
-
-                       /*
-                        * Copy over as many as possible extent records into
-                        * the previous page.
-                        */
-                       ext_diff = MIN(ext_avail, erp_next->er_extcount);
-                       memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
-                       erp->er_extcount += ext_diff;
-                       erp_next->er_extcount -= ext_diff;
-
-                       /*
-                        * If the next irec is empty now we can simply
-                        * remove it.
-                        */
-                       if (erp_next->er_extcount == 0) {
-                               /*
-                                * Free page before removing extent record
-                                * so er_extoffs don't get modified in
-                                * xfs_iext_irec_remove.
-                                */
-                               kmem_free(erp_next->er_extbuf);
-                               erp_next->er_extbuf = NULL;
-                               xfs_iext_irec_remove(ifp, erp_idx + 1);
-                               erp = &ifp->if_u1.if_ext_irec[erp_idx];
-                               nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
-
-                       /*
-                        * If the next irec is not empty move up the content
-                        * that has not been copied to the previous page to
-                        * the beggining of this one.
-                        */
-                       } else {
-                               memmove(erp_next->er_extbuf, &ep_next[ext_diff],
-                                       erp_next->er_extcount *
-                                       sizeof(xfs_bmbt_rec_t));
-                               ep_next = erp_next->er_extbuf;
-                               memset(&ep_next[erp_next->er_extcount], 0,
-                                       (XFS_LINEAR_EXTS -
-                                               erp_next->er_extcount) *
-                                       sizeof(xfs_bmbt_rec_t));
-                               erp_next->er_extoff += ext_diff;
-                       }
-               }
-
-               if (erp->er_extcount == XFS_LINEAR_EXTS) {
-                       erp_idx++;
-                       if (erp_idx < nlists)
-                               erp = &ifp->if_u1.if_ext_irec[erp_idx];
-                       else
-                               break;
-               }
-               ep = &erp->er_extbuf[erp->er_extcount];
-               erp_next = erp + 1;
-               ep_next = erp_next->er_extbuf;
-       }
-}
-
 /*
  * This is called to update the er_extoff field in the indirection
  * array when extents have been added or removed from one of the