[GFS2] gfs2_adjust_quota has broken unstuffing code
authorAbhijith Das <adas@redhat.com>
Thu, 6 Mar 2008 23:43:52 +0000 (17:43 -0600)
committerSteven Whitehouse <swhiteho@redhat.com>
Mon, 31 Mar 2008 09:41:30 +0000 (10:41 +0100)
This patch combines the 2 patches in bug 434736 to correct the lock
ordering in the unstuffing of the quota inode in gfs2_adjust_quota and
adjusting the number of revokes in gfs2_write_jdata_pagevec

Signed-off-by: Abhijith Das <adas@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
fs/gfs2/ops_address.c
fs/gfs2/quota.c

index 2483d8741060cd5dff6052b3625b1efca03614e0..e72fd47d71eb3ff84ea57f68b6bcae0e27b62148 100644 (file)
@@ -21,7 +21,6 @@
 #include <linux/gfs2_ondisk.h>
 #include <linux/lm_interface.h>
 #include <linux/backing-dev.h>
-#include <linux/pagevec.h>
 
 #include "gfs2.h"
 #include "incore.h"
@@ -278,7 +277,7 @@ static int gfs2_write_jdata_pagevec(struct address_space *mapping,
        int i;
        int ret;
 
-       ret = gfs2_trans_begin(sdp, nrblocks, 0);
+       ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
        if (ret < 0)
                return ret;
 
index 636bccfd2bcf4dfa27cf2b703a976f3ae5185dfe..c71f781db5d7952457ca39c920628b1d6228cf3b 100644 (file)
@@ -616,17 +616,9 @@ static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
        s64 value;
        int err = -EIO;
 
-       if (gfs2_is_stuffed(ip)) {
-               struct gfs2_alloc *al = gfs2_alloc_get(ip);
-               if (!al)
-                       return -ENOMEM;
-               /* just request 1 blk */
-               al->al_requested = 1;
-               gfs2_inplace_reserve(ip);
+       if (gfs2_is_stuffed(ip))
                gfs2_unstuff_dinode(ip, NULL);
-               gfs2_inplace_release(ip);
-               gfs2_alloc_put(ip);
-       }
+       
        page = grab_cache_page(mapping, index);
        if (!page)
                return -ENOMEM;
@@ -691,7 +683,7 @@ static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
        unsigned int qx, x;
        struct gfs2_quota_data *qd;
        loff_t offset;
-       unsigned int nalloc = 0;
+       unsigned int nalloc = 0, blocks;
        struct gfs2_alloc *al = NULL;
        int error;
 
@@ -728,34 +720,33 @@ static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
                        nalloc++;
        }
 
-       if (nalloc) {
-               al = gfs2_alloc_get(ip);
-               if (!al) {
-                       error = -ENOMEM;
-                       goto out_gunlock;
-               }
+       al = gfs2_alloc_get(ip);
+       if (!al) {
+               error = -ENOMEM;
+               goto out_gunlock;
+       }
+       /* 
+        * 1 blk for unstuffing inode if stuffed. We add this extra
+        * block to the reservation unconditionally. If the inode
+        * doesn't need unstuffing, the block will be released to the 
+        * rgrp since it won't be allocated during the transaction
+        */
+       al->al_requested = 1;
+       /* +1 in the end for block requested above for unstuffing */
+       blocks = num_qd * data_blocks + RES_DINODE + num_qd + 1;
 
-               al->al_requested = nalloc * (data_blocks + ind_blocks);
+       if (nalloc)
+               al->al_requested += nalloc * (data_blocks + ind_blocks);                
+       error = gfs2_inplace_reserve(ip);
+       if (error)
+               goto out_alloc;
 
-               error = gfs2_inplace_reserve(ip);
-               if (error)
-                       goto out_alloc;
-
-               error = gfs2_trans_begin(sdp,
-                                        al->al_rgd->rd_length +
-                                        num_qd * data_blocks +
-                                        nalloc * ind_blocks +
-                                        RES_DINODE + num_qd +
-                                        RES_STATFS, 0);
-               if (error)
-                       goto out_ipres;
-       } else {
-               error = gfs2_trans_begin(sdp,
-                                        num_qd * data_blocks +
-                                        RES_DINODE + num_qd, 0);
-               if (error)
-                       goto out_gunlock;
-       }
+       if (nalloc)
+               blocks += al->al_rgd->rd_length + nalloc * ind_blocks + RES_STATFS;
+
+       error = gfs2_trans_begin(sdp, blocks, 0);
+       if (error)
+               goto out_ipres;
 
        for (x = 0; x < num_qd; x++) {
                qd = qda[x];
@@ -774,11 +765,9 @@ static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
 out_end_trans:
        gfs2_trans_end(sdp);
 out_ipres:
-       if (nalloc)
-               gfs2_inplace_release(ip);
+       gfs2_inplace_release(ip);
 out_alloc:
-       if (nalloc)
-               gfs2_alloc_put(ip);
+       gfs2_alloc_put(ip);
 out_gunlock:
        gfs2_glock_dq_uninit(&i_gh);
 out: