[PATCH] ext3: ext3_symlink should use GFP_NOFS allocations inside (ver. 3)

[PATCH] ext3: ext3_symlink should use GFP_NOFS allocations inside (ver. 3)

[Kirill Korotaev]

[-- Attachment #1: Type: text/plain, Size: 666 bytes --]

Andrew,

Fixed both comments from Al Viro (thanks, Al):
- should have a separate helper
- should pass 0 instead of GFP_KERNEL in page_symlink()

This patch fixes illegal __GFP_FS allocation inside ext3
transaction in ext3_symlink().
Such allocation may re-enter ext3 code from
try_to_free_pages. But JBD/ext3 code keeps a pointer to current
journal handle in task_struct and, hence, is not reentrable.

This bug led to "Assertion failure in journal_dirty_metadata()" messages.

http://bugzilla.openvz.org/show_bug.cgi?id=115

Signed-Off-By: Andrey Savochkin <saw@saw.sw.com.sg>
Signed-Off-By: Kirill Korotaev <dev@openvz.org>

Thanks,
Kirill
P.S. against 2.6.16-rc5

[-- Attachment #2: diff-ext3-nofssymlink-20060210 --]
[-- Type: text/plain, Size: 2453 bytes --]

--- ./fs/ext3/namei.c.symlnkfix	2006-03-10 10:24:05.000000000 +0300
+++ ./fs/ext3/namei.c	2006-03-10 10:24:49.000000000 +0300
@@ -2141,7 +2141,7 @@ retry:
 		 * We have a transaction open.  All is sweetness.  It also sets
 		 * i_size in generic_commit_write().
 		 */
-		err = page_symlink(inode, symname, l);
+		err = __page_symlink(inode, symname, l, GFP_NOFS);
 		if (err) {
 			ext3_dec_count(handle, inode);
 			ext3_mark_inode_dirty(handle, inode);
--- ./fs/namei.c.symlnkfix	2006-03-10 10:24:05.000000000 +0300
+++ ./fs/namei.c	2006-03-10 10:34:58.000000000 +0300
@@ -2613,13 +2613,16 @@ void page_put_link(struct dentry *dentry
 	}
 }
 
-int page_symlink(struct inode *inode, const char *symname, int len)
+int __page_symlink(struct inode *inode, const char *symname, int len,
+		gfp_t gfp_mask)
 {
 	struct address_space *mapping = inode->i_mapping;
-	struct page *page = grab_cache_page(mapping, 0);
+	struct page *page;
 	int err = -ENOMEM;
 	char *kaddr;
 
+	page = find_or_create_page(mapping, 0,
+			mapping_gfp_mask(mapping) | gfp_mask);
 	if (!page)
 		goto fail;
 	err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
@@ -2654,6 +2657,11 @@ fail:
 	return err;
 }
 
+int page_symlink(struct inode *inode, const char *symname, int len)
+{
+	return __page_symlink(inode, symname, len, 0);
+}
+
 struct inode_operations page_symlink_inode_operations = {
 	.readlink	= generic_readlink,
 	.follow_link	= page_follow_link_light,
@@ -2672,6 +2680,7 @@ EXPORT_SYMBOL(lookup_one_len);
 EXPORT_SYMBOL(page_follow_link_light);
 EXPORT_SYMBOL(page_put_link);
 EXPORT_SYMBOL(page_readlink);
+EXPORT_SYMBOL(__page_symlink);
 EXPORT_SYMBOL(page_symlink);
 EXPORT_SYMBOL(page_symlink_inode_operations);
 EXPORT_SYMBOL(path_lookup);
--- ./include/linux/fs.h.symlnkfix	2006-03-10 10:24:05.000000000 +0300
+++ ./include/linux/fs.h	2006-03-10 10:27:40.000000000 +0300
@@ -1669,6 +1669,8 @@ extern int vfs_follow_link(struct nameid
 extern int page_readlink(struct dentry *, char __user *, int);
 extern void *page_follow_link_light(struct dentry *, struct nameidata *);
 extern void page_put_link(struct dentry *, struct nameidata *, void *);
+extern int __page_symlink(struct inode *inode, const char *symname, int len,
+		gfp_t gfp_mask);
 extern int page_symlink(struct inode *inode, const char *symname, int len);
 extern struct inode_operations page_symlink_inode_operations;
 extern int generic_readlink(struct dentry *, char __user *, int);

Re: [PATCH] ext3: ext3_symlink should use GFP_NOFS allocations inside (ver. 3)

[Arjan van de Ven]

On Fri, 2006-03-10 at 11:46 +0300, Kirill Korotaev wrote:
> Andrew,
> 
> Fixed both comments from Al Viro (thanks, Al):
> - should have a separate helper
> - should pass 0 instead of GFP_KERNEL in page_symlink()

>  
> +	page = find_or_create_page(mapping, 0,
> +			mapping_gfp_mask(mapping) | gfp_mask);



this does not work; GFP_NOFS has a bit *LESS* than GFP_KERNEL, not a bit
more. As such a | operation isn't going to be useful....

(So I think that while Al's intention was good, the implication of it
isn't ;)

Re: [PATCH] ext3: ext3_symlink should use GFP_NOFS allocations inside (ver. 3)

[Andrew Morton]

Arjan van de Ven <arjan@infradead.org> wrote:
>
> On Fri, 2006-03-10 at 11:46 +0300, Kirill Korotaev wrote:
> > Andrew,
> > 
> > Fixed both comments from Al Viro (thanks, Al):
> > - should have a separate helper
> > - should pass 0 instead of GFP_KERNEL in page_symlink()
> 
> >  
> > +	page = find_or_create_page(mapping, 0,
> > +			mapping_gfp_mask(mapping) | gfp_mask);
> 
> 
> 
> this does not work; GFP_NOFS has a bit *LESS* than GFP_KERNEL, not a bit
> more. As such a | operation isn't going to be useful....
> 
> (So I think that while Al's intention was good, the implication of it
> isn't ;)

Yup.  page_symlink() needs to pass in mapping_gfp_mask(inode->i_mapping)
and ext3 needs to pass in, umm,

	mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS

or

	GFP_NOFS|__GFP_HIGHMEM.

preferably the former I guess.

Re: [PATCH] ext3: ext3_symlink should use GFP_NOFS allocations inside (ver. 3)

[Kirill Korotaev]

>>Andrew,
>>
>>Fixed both comments from Al Viro (thanks, Al):
>>- should have a separate helper
>>- should pass 0 instead of GFP_KERNEL in page_symlink()
> 
> 
>> 
>>+	page = find_or_create_page(mapping, 0,
>>+			mapping_gfp_mask(mapping) | gfp_mask);
> 
> 
> 
> 
> this does not work; GFP_NOFS has a bit *LESS* than GFP_KERNEL, not a bit
> more. As such a | operation isn't going to be useful....
> 
> (So I think that while Al's intention was good, the implication of it
> isn't ;)
Oh no... Had to sleep well today... :/
Al, are you agree with the original patch then?
I don't think it is a good idea to introduce AND mask instead :)

Thanks,
Kirill

Re: [PATCH] ext3: ext3_symlink should use GFP_NOFS allocations inside (ver. 3)

[Al Viro]

On Fri, Mar 10, 2006 at 09:46:25AM +0100, Arjan van de Ven wrote:
> On Fri, 2006-03-10 at 11:46 +0300, Kirill Korotaev wrote:
> > Andrew,
> > 
> > Fixed both comments from Al Viro (thanks, Al):
> > - should have a separate helper
> > - should pass 0 instead of GFP_KERNEL in page_symlink()
> 
> >  
> > +	page = find_or_create_page(mapping, 0,
> > +			mapping_gfp_mask(mapping) | gfp_mask);
> 
> 
> 
> this does not work; GFP_NOFS has a bit *LESS* than GFP_KERNEL, not a bit
> more. As such a | operation isn't going to be useful....
> 
> (So I think that while Al's intention was good, the implication of it
> isn't ;)

s/|/^/ and accept my apologies...

Re: [PATCH] ext3: ext3_symlink should use GFP_NOFS allocations inside (ver. 3)

[Kirill Korotaev]

[-- Attachment #1: Type: text/plain, Size: 793 bytes --]

>>>Andrew,
>>>
>>>Fixed both comments from Al Viro (thanks, Al):
>>>- should have a separate helper
>>>- should pass 0 instead of GFP_KERNEL in page_symlink()
>>
>>> 
>>>+	page = find_or_create_page(mapping, 0,
>>>+			mapping_gfp_mask(mapping) | gfp_mask);
>>
>>
>>
>>this does not work; GFP_NOFS has a bit *LESS* than GFP_KERNEL, not a bit
>>more. As such a | operation isn't going to be useful....
>>
>>(So I think that while Al's intention was good, the implication of it
>>isn't ;)
> 
> 
> Yup.  page_symlink() needs to pass in mapping_gfp_mask(inode->i_mapping)
> and ext3 needs to pass in, umm,
> 
> 	mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS
> 
> or
> 
> 	GFP_NOFS|__GFP_HIGHMEM.
> 
> preferably the former I guess.

This looks reasonable.
See the patch attached.

Thanks,
Kirill


[-- Attachment #2: diff-ext3-nofssymlink-20060210 --]
[-- Type: text/plain, Size: 2501 bytes --]

--- ./fs/ext3/namei.c.symlnkfix	2006-03-10 10:24:05.000000000 +0300
+++ ./fs/ext3/namei.c	2006-03-10 12:06:00.000000000 +0300
@@ -2141,7 +2141,8 @@ retry:
 		 * We have a transaction open.  All is sweetness.  It also sets
 		 * i_size in generic_commit_write().
 		 */
-		err = page_symlink(inode, symname, l);
+		err = __page_symlink(inode, symname, l,
+				mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 		if (err) {
 			ext3_dec_count(handle, inode);
 			ext3_mark_inode_dirty(handle, inode);
--- ./fs/namei.c.symlnkfix	2006-03-10 10:24:05.000000000 +0300
+++ ./fs/namei.c	2006-03-10 12:07:47.000000000 +0300
@@ -2613,13 +2613,15 @@ void page_put_link(struct dentry *dentry
 	}
 }
 
-int page_symlink(struct inode *inode, const char *symname, int len)
+int __page_symlink(struct inode *inode, const char *symname, int len,
+		gfp_t gfp_mask)
 {
 	struct address_space *mapping = inode->i_mapping;
-	struct page *page = grab_cache_page(mapping, 0);
+	struct page *page;
 	int err = -ENOMEM;
 	char *kaddr;
 
+	page = find_or_create_page(mapping, 0, gfp_mask);
 	if (!page)
 		goto fail;
 	err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
@@ -2654,6 +2656,12 @@ fail:
 	return err;
 }
 
+int page_symlink(struct inode *inode, const char *symname, int len)
+{
+	return __page_symlink(inode, symname, len,
+			mapping_gfp_mask(inode->i_mapping));
+}
+
 struct inode_operations page_symlink_inode_operations = {
 	.readlink	= generic_readlink,
 	.follow_link	= page_follow_link_light,
@@ -2672,6 +2680,7 @@ EXPORT_SYMBOL(lookup_one_len);
 EXPORT_SYMBOL(page_follow_link_light);
 EXPORT_SYMBOL(page_put_link);
 EXPORT_SYMBOL(page_readlink);
+EXPORT_SYMBOL(__page_symlink);
 EXPORT_SYMBOL(page_symlink);
 EXPORT_SYMBOL(page_symlink_inode_operations);
 EXPORT_SYMBOL(path_lookup);
--- ./include/linux/fs.h.symlnkfix	2006-03-10 10:24:05.000000000 +0300
+++ ./include/linux/fs.h	2006-03-10 10:27:40.000000000 +0300
@@ -1669,6 +1669,8 @@ extern int vfs_follow_link(struct nameid
 extern int page_readlink(struct dentry *, char __user *, int);
 extern void *page_follow_link_light(struct dentry *, struct nameidata *);
 extern void page_put_link(struct dentry *, struct nameidata *, void *);
+extern int __page_symlink(struct inode *inode, const char *symname, int len,
+		gfp_t gfp_mask);
 extern int page_symlink(struct inode *inode, const char *symname, int len);
 extern struct inode_operations page_symlink_inode_operations;
 extern int generic_readlink(struct dentry *, char __user *, int);

Re: [PATCH] ext3: ext3_symlink should use GFP_NOFS allocations inside (ver. 3)

[Arjan van de Ven]

On Fri, 2006-03-10 at 08:59 +0000, Al Viro wrote:
> On Fri, Mar 10, 2006 at 09:46:25AM +0100, Arjan van de Ven wrote:
> > On Fri, 2006-03-10 at 11:46 +0300, Kirill Korotaev wrote:
> > > Andrew,
> > > 
> > > Fixed both comments from Al Viro (thanks, Al):
> > > - should have a separate helper
> > > - should pass 0 instead of GFP_KERNEL in page_symlink()
> > 
> > >  
> > > +	page = find_or_create_page(mapping, 0,
> > > +			mapping_gfp_mask(mapping) | gfp_mask);
> > 
> > 
> > 
> > this does not work; GFP_NOFS has a bit *LESS* than GFP_KERNEL, not a bit
> > more. As such a | operation isn't going to be useful....
> > 
> > (So I think that while Al's intention was good, the implication of it
> > isn't ;)
> 
> s/|/^/ and accept my apologies...

I think you mean
& ~() 

since xor... can flip it on if it was off after all