]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
ufs_truncate_blocks(): fix the case when size is in the last direct block
authorAl Viro <viro@zeniv.linux.org.uk>
Thu, 15 Jun 2017 07:57:46 +0000 (03:57 -0400)
committerAl Viro <viro@zeniv.linux.org.uk>
Thu, 15 Jun 2017 07:57:46 +0000 (03:57 -0400)
The logics when deciding whether we need to do anything with direct blocks
is broken when new size is within the last direct block.  It's better to
find the path to the last byte _not_ to be removed and use that instead
of the path to the beginning of the first block to be freed...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/ufs/inode.c

index 1dda6c4875f9fd5f44bc1bd218e16da8028ee56b..9f4590261134085cbabcea1ee9a72382239ba479 100644 (file)
@@ -886,7 +886,6 @@ static inline void free_data(struct to_free *ctx, u64 from, unsigned count)
        ctx->to = from + count;
 }
 
-#define DIRECT_BLOCK ((inode->i_size + uspi->s_bsize - 1) >> uspi->s_bshift)
 #define DIRECT_FRAGMENT ((inode->i_size + uspi->s_fsize - 1) >> uspi->s_fshift)
 
 static void ufs_trunc_direct(struct inode *inode)
@@ -1124,19 +1123,24 @@ static void ufs_truncate_blocks(struct inode *inode)
        struct super_block *sb = inode->i_sb;
        struct ufs_sb_private_info *uspi = UFS_SB(sb)->s_uspi;
        unsigned offsets[4];
-       int depth = ufs_block_to_path(inode, DIRECT_BLOCK, offsets);
+       int depth;
        int depth2;
        unsigned i;
        struct ufs_buffer_head *ubh[3];
        void *p;
        u64 block;
 
-       if (!depth)
-               return;
+       if (inode->i_size) {
+               sector_t last = (inode->i_size - 1) >> uspi->s_bshift;
+               depth = ufs_block_to_path(inode, last, offsets);
+               if (!depth)
+                       return;
+       } else {
+               depth = 1;
+       }
 
-       /* find the last non-zero in offsets[] */
        for (depth2 = depth - 1; depth2; depth2--)
-               if (offsets[depth2])
+               if (offsets[depth2] != uspi->s_apb - 1)
                        break;
 
        mutex_lock(&ufsi->truncate_mutex);
@@ -1145,9 +1149,8 @@ static void ufs_truncate_blocks(struct inode *inode)
                offsets[0] = UFS_IND_BLOCK;
        } else {
                /* get the blocks that should be partially emptied */
-               p = ufs_get_direct_data_ptr(uspi, ufsi, offsets[0]);
+               p = ufs_get_direct_data_ptr(uspi, ufsi, offsets[0]++);
                for (i = 0; i < depth2; i++) {
-                       offsets[i]++;   /* next branch is fully freed */
                        block = ufs_data_ptr_to_cpu(sb, p);
                        if (!block)
                                break;
@@ -1158,7 +1161,7 @@ static void ufs_truncate_blocks(struct inode *inode)
                                write_sequnlock(&ufsi->meta_lock);
                                break;
                        }
-                       p = ubh_get_data_ptr(uspi, ubh[i], offsets[i + 1]);
+                       p = ubh_get_data_ptr(uspi, ubh[i], offsets[i + 1]++);
                }
                while (i--)
                        free_branch_tail(inode, offsets[i + 1], ubh[i], depth - i - 1);