]> git.kernelconcepts.de Git - karo-tx-linux.git/blobdiff - fs/xfs/xfs_bmap_util.c
Merge tag 'master-2014-12-01' of git://git.kernel.org/pub/scm/linux/kernel/git/linvil...
[karo-tx-linux.git] / fs / xfs / xfs_bmap_util.c
index 92e8f99a58575b0667be026c9bfc6da5eeeae76d..281002689d64f858c605ce4a8d2aa72864034549 100644 (file)
@@ -1338,7 +1338,10 @@ xfs_free_file_space(
        goto out;
 }
 
-
+/*
+ * Preallocate and zero a range of a file. This mechanism has the allocation
+ * semantics of fallocate and in addition converts data in the range to zeroes.
+ */
 int
 xfs_zero_file_space(
        struct xfs_inode        *ip,
@@ -1346,65 +1349,30 @@ xfs_zero_file_space(
        xfs_off_t               len)
 {
        struct xfs_mount        *mp = ip->i_mount;
-       uint                    granularity;
-       xfs_off_t               start_boundary;
-       xfs_off_t               end_boundary;
+       uint                    blksize;
        int                     error;
 
        trace_xfs_zero_file_space(ip);
 
-       granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
+       blksize = 1 << mp->m_sb.sb_blocklog;
 
        /*
-        * Round the range of extents we are going to convert inwards.  If the
-        * offset is aligned, then it doesn't get changed so we zero from the
-        * start of the block offset points to.
+        * Punch a hole and prealloc the range. We use hole punch rather than
+        * unwritten extent conversion for two reasons:
+        *
+        * 1.) Hole punch handles partial block zeroing for us.
+        *
+        * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
+        * by virtue of the hole punch.
         */
-       start_boundary = round_up(offset, granularity);
-       end_boundary = round_down(offset + len, granularity);
-
-       ASSERT(start_boundary >= offset);
-       ASSERT(end_boundary <= offset + len);
-
-       if (start_boundary < end_boundary - 1) {
-               /*
-                * Writeback the range to ensure any inode size updates due to
-                * appending writes make it to disk (otherwise we could just
-                * punch out the delalloc blocks).
-                */
-               error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
-                               start_boundary, end_boundary - 1);
-               if (error)
-                       goto out;
-               truncate_pagecache_range(VFS_I(ip), start_boundary,
-                                        end_boundary - 1);
-
-               /* convert the blocks */
-               error = xfs_alloc_file_space(ip, start_boundary,
-                                       end_boundary - start_boundary - 1,
-                                       XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT);
-               if (error)
-                       goto out;
-
-               /* We've handled the interior of the range, now for the edges */
-               if (start_boundary != offset) {
-                       error = xfs_iozero(ip, offset, start_boundary - offset);
-                       if (error)
-                               goto out;
-               }
-
-               if (end_boundary != offset + len)
-                       error = xfs_iozero(ip, end_boundary,
-                                          offset + len - end_boundary);
-
-       } else {
-               /*
-                * It's either a sub-granularity range or the range spanned lies
-                * partially across two adjacent blocks.
-                */
-               error = xfs_iozero(ip, offset, len);
-       }
+       error = xfs_free_file_space(ip, offset, len);
+       if (error)
+               goto out;
 
+       error = xfs_alloc_file_space(ip, round_down(offset, blksize),
+                                    round_up(offset + len, blksize) -
+                                    round_down(offset, blksize),
+                                    XFS_BMAPI_PREALLOC);
 out:
        return error;