Reiser4 bug fixes

PATCH In review Hans Reiser 2 participants 2 messages Jul 19, 2006 – Aug 19, 2006

Share & tools

Hans Reiser

author

2006-07-19 20:22 UTC #0 lore raw

PATCH 1/1 Reiser4 bug fixes

Hopefully this will make reiser4 stable code again. Or at least, a lot more stable than v3 (or other FSes) was when it went in....

Signed-off-by Hans Reiser

From: Vladimir Saveliev <[email protected]>

This patch contains 4 reiser4 bug fixes:
- missing long term lock unlock on read' error handling code path is added
- truncate_inode_pages is now called with embedded mapping.
It is needed to deal properly with block device special files.
- copy_to_user might be called for more bytes than was prefaulted with fault_in_pages_writeable. That caused undesirable major page faults.
- readdir bug similar to the previous one:
call to filldir might lead to deadlock due to major page fault.

Signed-off-by Vladimir Saveliev <[email protected]>

fs/reiser4/super_ops.c renamed from fs/reiser4/super_ops.c~reiser4-one-line-fixes +1 -1

diff -puN fs/reiser4/super_ops.c~reiser4-one-line-fixes fs/reiser4/super_ops.c
@@ -202,7 +202,7 @@ static void reiser4_delete_inode(struct 
 			fplug->delete_object(inode);
 	}
 
-	truncate_inode_pages(inode->i_mapping, 0);
+	truncate_inode_pages(&inode->i_data, 0);
 	inode->i_blocks = 0;
 	clear_inode(inode);
 	reiser4_exit_context(ctx);

fs/reiser4/plugin/file/file.c renamed from fs/reiser4/plugin/file/file.c~reiser4-one-line-fixes +16 -5

diff -puN fs/reiser4/plugin/file/file.c~reiser4-one-line-fixes fs/reiser4/plugin/file/file.c
@@ -781,9 +781,12 @@ int find_or_create_extent(struct page *p
 
 	lock_page(page);
 	node = jnode_of_page(page);
-	unlock_page(page);
-	if (IS_ERR(node))
+	if (IS_ERR(node)) {
+		unlock_page(page);
 		return PTR_ERR(node);
+	}
+	JF_SET(node, JNODE_WRITE_PREPARED);
+	unlock_page(page);
 
 	if (node->blocknr == 0) {
 		plugged_hole = 0;

@@ -791,6 +794,7 @@ int find_or_create_extent(struct page *p
 				       (loff_t)page->index << PAGE_CACHE_SHIFT,
 				       &plugged_hole);
 		if (result) {
+ 			JF_CLR(node, JNODE_WRITE_PREPARED);
 			jput(node);
 			warning("", "update_extent failed: %d", result);
 			return result;

@@ -806,6 +810,7 @@ int find_or_create_extent(struct page *p
 	}
 
 	BUG_ON(node->atom == NULL);
+	JF_CLR(node, JNODE_WRITE_PREPARED);
 	jput(node);
 
 	if (get_current_context()->entd) {

@@ -1631,14 +1636,18 @@ static size_t read_file(hint_t * hint, s
 			/* error happened */
 			break;
 
-		if (coord->between != AT_UNIT)
+		if (coord->between != AT_UNIT) {
 			/* there were no items corresponding to given offset */
+			done_lh(hint->ext_coord.lh);
 			break;
+		}
 
 		loaded = coord->node;
 		result = zload(loaded);
-		if (unlikely(result))
+		if (unlikely(result)) {
+			done_lh(hint->ext_coord.lh);
 			break;
+		}
 
 		if (hint->ext_coord.valid == 0)
 			validate_extended_coord(&hint->ext_coord,

@@ -1729,7 +1738,9 @@ ssize_t read_unix_file(struct file *file
 			return RETERR(-EFAULT);
 		}
 
-		read = read_file(hint, file, buf, left, off);
+		read = read_file(hint, file, buf,
+				 left > PAGE_CACHE_SIZE ? PAGE_CACHE_SIZE : left,
+				 off);
 
  		drop_nonexclusive_access(uf_info);

fs/reiser4/plugin/file_ops_readdir.c renamed from fs/reiser4/plugin/file_ops_readdir.c~reiser4-one-line-fixes +2 -1

diff -puN fs/reiser4/plugin/file_ops_readdir.c~reiser4-one-line-fixes fs/reiser4/plugin/file_ops_readdir.c
@@ -349,6 +349,7 @@ feed_entry(struct file *f,
 	 */
 	assert("nikita-3436", lock_stack_isclean(get_current_lock_stack()));
 
+	txn_restart_current();
 	result = filldir(dirent, name, (int)strlen(name),
 			 /* offset of this entry */
 			 f->f_pos,

@@ -630,7 +631,7 @@ int readdir_common(struct file *f /* dir
 	detach_fsdata(f);
 
 	/* try to update directory's atime */
-	if (reiser4_grab_space(inode_file_plugin(inode)->estimate.update(inode),
+	if (reiser4_grab_space_force(inode_file_plugin(inode)->estimate.update(inode),
 			       BA_CAN_COMMIT) != 0)
 		warning("", "failed to update atime on readdir: %llu",
 			get_inode_oid(inode));

Christoph Hellwig

2006-08-19 20:16 UTC #1 lore raw

Quoted 24 lines · Hans Reiser

Hopefully this will make reiser4 stable code again. Or at least, a lot more stable than v3 (or other FSes) was when it went in....
Signed-off-by Hans Reiser
From: Vladimir Saveliev <[email protected]>
This patch contains 4 reiser4 bug fixes:
- missing long term lock unlock on read' error handling code path is added
- truncate_inode_pages is now called with embedded mapping.
It is needed to deal properly with block device special files.
- copy_to_user might be called for more bytes than was prefaulted with fault_in_pages_writeable. That caused undesirable major page faults.
- readdir bug similar to the previous one:
call to filldir might lead to deadlock due to major page fault.
Signed-off-by Vladimir Saveliev <[email protected]>
diff -puN fs/reiser4/super_ops.c~reiser4-one-line-fixes fs/reiser4/super_ops.c

fs/reiser4/super_ops.c static void reiser4_delete_inode(struct · line 205

Quoted 5 lines

fs/reiser4/super_ops.c renamed from fs/reiser4/super_ops.c~reiser4-one-line-fixes +1 -1

@@ -202,7 +202,7 @@ static void reiser4_delete_inode(struct 
 			fplug->delete_object(inode);
 	}
 
-	truncate_inode_pages(inode->i_mapping, 0);
+	truncate_inode_pages(&inode->i_data, 0);

Where do you reset i_mapping to be different from i_data? It is a valid thing to do in theory (that's why we have those different fields), but in practice most usages I've seen are bogus.

(And yes, this is not a personal attack again you, just a normal review)