Skip to contentSkip to conversation

overlayfs: Pass O_TRUNC flag to underlying filesystem

PATCH In review Vivek Goyal 2 participants 5 messages Apr 21, 2020 – Apr 22, 2020 Overlay filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 overlayfs: Pass O_TRUNC flag to underlying filesystem

As of now during open(), we don't pass bunch of flags to underlying filesystem. O_TRUNC is one of these. Normally this is not a problem as VFS calls ->setattr() with zero size and underlying filesystem sets file size to 0.

But when overlayfs is running on top of virtiofs, it has an optimization where it does not send setattr request to server if dectects that truncation is part of open(O_TRUNC). It assumes that server already zeroed file size as part of open(O_TRUNC).

fuse_do_setattr() {
	if (attr->ia_valid & ATTR_OPEN) {
		/*
		 * No need to send request to userspace, since actual
		 * truncation has already been done by OPEN.  But still
		 * need to truncate page cache.
		 */
	}
}

IOW, fuse expects O_TRUNC to be passed to it as part of open flags.

But currently overlayfs does not pass O_TRUNC to underlying filesystem hence fuse/virtiofs breaks. Setup overlayfs on top of virtiofs and following does not zero the file size of a file is either upper only or has already been copied up.

fd = open(foo.txt, O_TRUNC | O_WRONLY);

Fix it by passing O_TRUNC to underlying filesystem.

I found this problem while running unionmount-testsuite. It fails.

***
*** ./run --ov --samefs --ts=0 open-trunc
***
TEST open-trunc.py:10: Open O_TRUNC|O_RDONLY
./run --open-file /mnt/virtiofs/mnt/a/foo100 -r -t -R ./run --open-file /mnt/virtiofs/mnt/a/foo100 -r -t -R
TEST open-trunc.py:18: Open O_TRUNC|O_WRONLY
./run --open-file /mnt/virtiofs/mnt/a/foo101 -w -t -W q ./run --open-file /mnt/virtiofs/mnt/a/foo101 -r -R q ./run --open-file /mnt/virtiofs/mnt/a/foo101 -w -t -W p ./run --open-file /mnt/virtiofs/mnt/a/foo101 -r -R p
TEST open-trunc.py:28: Open O_TRUNC|O_APPEND|O_WRONLY
./run --open-file /mnt/virtiofs/mnt/a/foo102 -a -t -W q ./run --open-file /mnt/virtiofs/mnt/a/foo102 -r -R q ./run --open-file /mnt/virtiofs/mnt/a/foo102 -a -t -W p ./run --open-file /mnt/virtiofs/mnt/a/foo102 -r -R p
/mnt/virtiofs/mnt/a/foo102: File size wrong (got 2, want 1)

After this patch, unionmount-testsuite passes with overlayfs on top of virtiofs.

---
 fs/overlayfs/file.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/overlayfs/file.c +1 -1
Index: redhat-linux/fs/overlayfs/file.c
===================================================================
@@ -134,7 +134,7 @@ static int ovl_open(struct inode *inode,
 		return err;
 
 	/* No longer need these flags, so don't pass them on to underlying fs */
-	file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
+	file->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY);
 
 	realfile = ovl_open_realfile(file, ovl_inode_realdata(inode));
 	if (IS_ERR(realfile))
maintainer
#1 lore raw
Quoted 23 lines · Vivek Goyal

As of now during open(), we don't pass bunch of flags to underlying filesystem. O_TRUNC is one of these. Normally this is not a problem as VFS calls ->setattr() with zero size and underlying filesystem sets file size to 0.

But when overlayfs is running on top of virtiofs, it has an optimization where it does not send setattr request to server if dectects that truncation is part of open(O_TRUNC). It assumes that server already zeroed file size as part of open(O_TRUNC).

fuse_do_setattr() {
        if (attr->ia_valid & ATTR_OPEN) {
                /*
                 * No need to send request to userspace, since actual
                 * truncation has already been done by OPEN.  But still
                 * need to truncate page cache.
                 */
        }
}

IOW, fuse expects O_TRUNC to be passed to it as part of open flags.

Quoted 8 lines

But currently overlayfs does not pass O_TRUNC to underlying filesystem hence fuse/virtiofs breaks. Setup overlayfs on top of virtiofs and following does not zero the file size of a file is either upper only or has already been copied up.

fd = open(foo.txt, O_TRUNC | O_WRONLY);

Fix it by passing O_TRUNC to underlying filesystem.

Or clear ATTR_OPEN in ovl_setattr()

Need to think about side effects of passing O_TRUNC down to underlying
fs. Clearing ATTR_OPEN seems obviously safe, so as a quick fix I'd
rather go with that for now.

Thanks,
Miklos

#2 lore raw
Quoted 31 lines · Miklos Szeredi
Quoted 30 lines · Vivek Goyal

As of now during open(), we don't pass bunch of flags to underlying filesystem. O_TRUNC is one of these. Normally this is not a problem as VFS calls ->setattr() with zero size and underlying filesystem sets file size to 0.

But when overlayfs is running on top of virtiofs, it has an optimization where it does not send setattr request to server if dectects that truncation is part of open(O_TRUNC). It assumes that server already zeroed file size as part of open(O_TRUNC).

fuse_do_setattr() {
        if (attr->ia_valid & ATTR_OPEN) {
                /*
                 * No need to send request to userspace, since actual
                 * truncation has already been done by OPEN.  But still
                 * need to truncate page cache.
                 */
        }
}

IOW, fuse expects O_TRUNC to be passed to it as part of open flags.

But currently overlayfs does not pass O_TRUNC to underlying filesystem hence fuse/virtiofs breaks. Setup overlayfs on top of virtiofs and following does not zero the file size of a file is either upper only or has already been copied up.

fd = open(foo.txt, O_TRUNC | O_WRONLY);

Quoted 8 lines
Quoted 1 line

Fix it by passing O_TRUNC to underlying filesystem.

Or clear ATTR_OPEN in ovl_setattr()

Need to think about side effects of passing O_TRUNC down to underlying
fs. Clearing ATTR_OPEN seems obviously safe, so as a quick fix I'd
rather go with that for now.

Ok, I will send another patch which clears ATTR_OPEN in ovl_setattr().

Thanks
Vivek

#3 lore raw
In reply to Miklos Szeredi
Quoted 31 lines · Miklos Szeredi
Quoted 30 lines · Vivek Goyal

As of now during open(), we don't pass bunch of flags to underlying filesystem. O_TRUNC is one of these. Normally this is not a problem as VFS calls ->setattr() with zero size and underlying filesystem sets file size to 0.

But when overlayfs is running on top of virtiofs, it has an optimization where it does not send setattr request to server if dectects that truncation is part of open(O_TRUNC). It assumes that server already zeroed file size as part of open(O_TRUNC).

fuse_do_setattr() {
        if (attr->ia_valid & ATTR_OPEN) {
                /*
                 * No need to send request to userspace, since actual
                 * truncation has already been done by OPEN.  But still
                 * need to truncate page cache.
                 */
        }
}

IOW, fuse expects O_TRUNC to be passed to it as part of open flags.

But currently overlayfs does not pass O_TRUNC to underlying filesystem hence fuse/virtiofs breaks. Setup overlayfs on top of virtiofs and following does not zero the file size of a file is either upper only or has already been copied up.

fd = open(foo.txt, O_TRUNC | O_WRONLY);

Quoted 8 lines
Quoted 1 line

Fix it by passing O_TRUNC to underlying filesystem.

Or clear ATTR_OPEN in ovl_setattr()

Need to think about side effects of passing O_TRUNC down to underlying
fs. Clearing ATTR_OPEN seems obviously safe, so as a quick fix I'd
rather go with that for now.

Found another interesting problem while I cleared ATTR_OPEN. VFS also sets ATTR_FILE and attr->ia_file has ovl file pointer. ovl_setattr() does not look at this attribute and passes it to underlying layer as it is. Fuse thinks it got a valid file object and passes file handle to server and server complains -EBADF.

ext4/xfs don't seem to look at ATTR_FILE, so it did not create problems so far.

For now, I will simply reset ATTR_FILE, indicating to lower layers don't use attr->ia_file.

Thanks
Vivek

maintainer
#4 lore raw
Quoted 48 lines · Vivek Goyal
Quoted 31 lines · Miklos Szeredi
Quoted 30 lines · Vivek Goyal

As of now during open(), we don't pass bunch of flags to underlying filesystem. O_TRUNC is one of these. Normally this is not a problem as VFS calls ->setattr() with zero size and underlying filesystem sets file size to 0.

But when overlayfs is running on top of virtiofs, it has an optimization where it does not send setattr request to server if dectects that truncation is part of open(O_TRUNC). It assumes that server already zeroed file size as part of open(O_TRUNC).

fuse_do_setattr() {
        if (attr->ia_valid & ATTR_OPEN) {
                /*
                 * No need to send request to userspace, since actual
                 * truncation has already been done by OPEN.  But still
                 * need to truncate page cache.
                 */
        }
}

IOW, fuse expects O_TRUNC to be passed to it as part of open flags.

But currently overlayfs does not pass O_TRUNC to underlying filesystem hence fuse/virtiofs breaks. Setup overlayfs on top of virtiofs and following does not zero the file size of a file is either upper only or has already been copied up.

fd = open(foo.txt, O_TRUNC | O_WRONLY);

Quoted 8 lines
Quoted 1 line

Fix it by passing O_TRUNC to underlying filesystem.

Or clear ATTR_OPEN in ovl_setattr()

Need to think about side effects of passing O_TRUNC down to underlying
fs. Clearing ATTR_OPEN seems obviously safe, so as a quick fix I'd
rather go with that for now.

Found another interesting problem while I cleared ATTR_OPEN. VFS also sets ATTR_FILE and attr->ia_file has ovl file pointer. ovl_setattr() does not look at this attribute and passes it to underlying layer as it is. Fuse thinks it got a valid file object and passes file handle to server and server complains -EBADF.

Quoted 5 lines

ext4/xfs don't seem to look at ATTR_FILE, so it did not create problems so far.

For now, I will simply reset ATTR_FILE, indicating to lower layers don't use attr->ia_file.

Right.

The solution to that would be to replace attr->ia_file with the real file, but that can wait until we see any use case for that.

Thanks,
Miklos