]> rtime.felk.cvut.cz Git - mcf548x/linux.git/commitdiff
xfs: new truncate sequence
authorChristoph Hellwig <hch@infradead.org>
Mon, 14 Jun 2010 09:17:31 +0000 (05:17 -0400)
committerAl Viro <viro@zeniv.linux.org.uk>
Mon, 9 Aug 2010 20:47:42 +0000 (16:47 -0400)
Convert XFS to the new truncate sequence.  We still can have errors after
updating the file size in xfs_setattr, but these are real I/O errors and lead
to a transaction abort and filesystem shutdown, so they are not an issue.

Errors from ->write_begin and write_end can now be handled correctly because
we can actually get rid of the delalloc extents while previous the buffer
state was stipped in block_invalidatepage.

There is still no error handling for ->direct_IO, because doing so will need
some major restructuring given that we only have the iolock shared and do not
hold i_mutex at all.  Fortunately leaving the normally allocated blocks behind
there is not a major issue and this will get cleaned up by xfs_free_eofblock
later.

Note: the patch is against Al's vfs.git tree as that contains the nessecary
preparations.  I'd prefer to get it applied there so that we can get some
testing in linux-next.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/xfs/linux-2.6/xfs_aops.c
fs/xfs/linux-2.6/xfs_iops.c
fs/xfs/linux-2.6/xfs_linux.h
fs/xfs/xfs_vnodeops.c

index bf7aad0d78b899667061abd59960b3955ba50982..15412fe15c3a47f3cc744af36cecfe2b01e5a5a3 100644 (file)
@@ -1494,6 +1494,22 @@ xfs_vm_direct_IO(
        return ret;
 }
 
+STATIC void
+xfs_vm_write_failed(
+       struct address_space    *mapping,
+       loff_t                  to)
+{
+       struct inode            *inode = mapping->host;
+
+       if (to > inode->i_size) {
+               struct iattr    ia = {
+                       .ia_valid       = ATTR_SIZE | ATTR_FORCE,
+                       .ia_size        = inode->i_size,
+               };
+               xfs_setattr(XFS_I(inode), &ia, XFS_ATTR_NOLOCK);
+       }
+}
+
 STATIC int
 xfs_vm_write_begin(
        struct file             *file,
@@ -1508,12 +1524,26 @@ xfs_vm_write_begin(
 
        ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS,
                                pagep, xfs_get_blocks);
-       if (unlikely(ret)) {
-               loff_t isize = mapping->host->i_size;
-               if (pos + len > isize)
-                       vmtruncate(mapping->host, isize);
-       }
+       if (unlikely(ret))
+               xfs_vm_write_failed(mapping, pos + len);
+       return ret;
+}
+
+STATIC int
+xfs_vm_write_end(
+       struct file             *file,
+       struct address_space    *mapping,
+       loff_t                  pos,
+       unsigned                len,
+       unsigned                copied,
+       struct page             *page,
+       void                    *fsdata)
+{
+       int                     ret;
 
+       ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
+       if (unlikely(ret < len))
+               xfs_vm_write_failed(mapping, pos + len);
        return ret;
 }
 
@@ -1559,7 +1589,7 @@ const struct address_space_operations xfs_address_space_operations = {
        .releasepage            = xfs_vm_releasepage,
        .invalidatepage         = xfs_vm_invalidatepage,
        .write_begin            = xfs_vm_write_begin,
-       .write_end              = generic_write_end,
+       .write_end              = xfs_vm_write_end,
        .bmap                   = xfs_vm_bmap,
        .direct_IO              = xfs_vm_direct_IO,
        .migratepage            = buffer_migrate_page,
index 536b81e63a3d8e397d8e487192f938809ee26923..62dd349faceebce88741d911109a935500c43757 100644 (file)
@@ -540,21 +540,6 @@ xfs_vn_setattr(
        return -xfs_setattr(XFS_I(dentry->d_inode), iattr, 0);
 }
 
-/*
- * block_truncate_page can return an error, but we can't propagate it
- * at all here. Leave a complaint + stack trace in the syslog because
- * this could be bad. If it is bad, we need to propagate the error further.
- */
-STATIC void
-xfs_vn_truncate(
-       struct inode    *inode)
-{
-       int     error;
-       error = block_truncate_page(inode->i_mapping, inode->i_size,
-                                                       xfs_get_blocks);
-       WARN_ON(error);
-}
-
 STATIC long
 xfs_vn_fallocate(
        struct inode    *inode,
@@ -694,7 +679,6 @@ xfs_vn_fiemap(
 
 static const struct inode_operations xfs_inode_operations = {
        .check_acl              = xfs_check_acl,
-       .truncate               = xfs_vn_truncate,
        .getattr                = xfs_vn_getattr,
        .setattr                = xfs_vn_setattr,
        .setxattr               = generic_setxattr,
index 998a9d7fb9c8964d2174f62f935f6fd46b4f2042..2fa0bd9ebc7f152e7f81f82d07fc6dc665d15db2 100644 (file)
  */
 #define xfs_sort(a,n,s,fn)     sort(a,n,s,fn,NULL)
 #define xfs_stack_trace()      dump_stack()
-#define xfs_itruncate_data(ip, off)    \
-       (-vmtruncate(VFS_I(ip), (off)))
 
 
 /* Move the kernel do_div definition off to one side */
index 3ac137dd531bb5481b5ec04ef085de3a6506878d..66d585c6917cfd55e50447b8048a2adc3f60d68c 100644 (file)
@@ -221,8 +221,11 @@ xfs_setattr(
                         * transaction to modify the i_size.
                         */
                        code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
+                       if (code)
+                               goto error_return;
                }
                xfs_iunlock(ip, XFS_ILOCK_EXCL);
+               lock_flags &= ~XFS_ILOCK_EXCL;
 
                /*
                 * We are going to log the inode size change in this
@@ -236,36 +239,35 @@ xfs_setattr(
                 * really care about here and prevents waiting for other data
                 * not within the range we care about here.
                 */
-               if (!code &&
-                   ip->i_size != ip->i_d.di_size &&
+               if (ip->i_size != ip->i_d.di_size &&
                    iattr->ia_size > ip->i_d.di_size) {
                        code = xfs_flush_pages(ip,
                                        ip->i_d.di_size, iattr->ia_size,
                                        XBF_ASYNC, FI_NONE);
+                       if (code)
+                               goto error_return;
                }
 
                /* wait for all I/O to complete */
                xfs_ioend_wait(ip);
 
-               if (!code)
-                       code = xfs_itruncate_data(ip, iattr->ia_size);
-               if (code) {
-                       ASSERT(tp == NULL);
-                       lock_flags &= ~XFS_ILOCK_EXCL;
-                       ASSERT(lock_flags == XFS_IOLOCK_EXCL || !need_iolock);
+               code = -block_truncate_page(inode->i_mapping, iattr->ia_size,
+                                           xfs_get_blocks);
+               if (code)
                        goto error_return;
-               }
+
                tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
-               if ((code = xfs_trans_reserve(tp, 0,
-                                            XFS_ITRUNCATE_LOG_RES(mp), 0,
-                                            XFS_TRANS_PERM_LOG_RES,
-                                            XFS_ITRUNCATE_LOG_COUNT))) {
-                       xfs_trans_cancel(tp, 0);
-                       if (need_iolock)
-                               xfs_iunlock(ip, XFS_IOLOCK_EXCL);
-                       return code;
-               }
+               code = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
+                                        XFS_TRANS_PERM_LOG_RES,
+                                        XFS_ITRUNCATE_LOG_COUNT);
+               if (code)
+                       goto error_return;
+
+               truncate_setsize(inode, iattr->ia_size);
+
                commit_flags = XFS_TRANS_RELEASE_LOG_RES;
+               lock_flags |= XFS_ILOCK_EXCL;
+
                xfs_ilock(ip, XFS_ILOCK_EXCL);
 
                xfs_trans_ijoin(tp, ip);