Skip to contentSkip to conversation

cifs: create a helper to find a writeable handle by path name

PATCH In review Ronnie Sahlberg 3 participants 3 messages Aug 29, 2019 – Sep 11, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 cifs: create a helper to find a writeable handle by path name

rename() takes a path for old_file and in SMB2 we used to just create a compound for create(old_path)/rename/close().
If we already have a writable handle we can avoid the create() and close() altogether and just use the existing handle.

For this situation, as we avoid doing the create()
we also avoid triggering an oplock break for the existing handle.

---
 fs/cifs/cifsproto.h |  2 ++
 fs/cifs/dir.c       |  2 +-
 fs/cifs/file.c      | 35 ++++++++++++++++++++
 fs/cifs/smb2inode.c | 94 ++++++++++++++++++++++++++++++++++++++---------------
 4 files changed, 106 insertions(+), 27 deletions(-)
fs/cifs/cifsproto.h +2 -0
index 592a6cea2b79..be206744407c 100644
@@ -137,6 +137,8 @@ extern struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *, bool);
 extern int cifs_get_writable_file(struct cifsInodeInfo *cifs_inode,
 				  bool fsuid_only,
 				  struct cifsFileInfo **ret_file);
+extern int cifs_get_writable_path(struct cifs_tcon *tcon, const char *name,
+				  struct cifsFileInfo **ret_file);
 extern struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *, bool);
 extern unsigned int smbCalcSize(void *buf, struct TCP_Server_Info *server);
 extern int decode_negTokenInit(unsigned char *security_blob, int length,
fs/cifs/dir.c +1 -1
index be424e81e3ad..dd5ac841aefa 100644
@@ -125,7 +125,7 @@ build_path_from_dentry_optional_prefix(struct dentry *direntry, bool prefix)
 	}
 	rcu_read_unlock();
 
-	full_path = kmalloc(namelen+1, GFP_KERNEL);
+	full_path = kmalloc(namelen+1, GFP_ATOMIC);
 	if (full_path == NULL)
 		return full_path;
 	full_path[namelen] = 0;	/* trailing null */
fs/cifs/file.c +35 -0
index 97090693d182..885740496dae 100644
@@ -1980,6 +1980,41 @@ find_writable_file(struct cifsInodeInfo *cifs_inode, bool fsuid_only)
 	return cfile;
 }
 
+int
+cifs_get_writable_path(struct cifs_tcon *tcon, const char *name,
+		       struct cifsFileInfo **ret_file)
+{
+	struct list_head *tmp;
+	struct cifsFileInfo *cfile;
+	struct cifsInodeInfo *cinode;
+	char *full_path;
+
+	*ret_file = NULL;
+
+	spin_lock(&tcon->open_file_lock);
+	list_for_each(tmp, &tcon->openFileList) {
+		cfile = list_entry(tmp, struct cifsFileInfo,
+			     tlist);
+		full_path = build_path_from_dentry(cfile->dentry);
+		if (full_path == NULL) {
+			spin_unlock(&tcon->open_file_lock);
+			return -ENOMEM;
+		}
+		if (strcmp(full_path, name)) {
+			kfree(full_path);
+			continue;
+		}
+
+		kfree(full_path);
+		cinode = CIFS_I(d_inode(cfile->dentry));
+		spin_unlock(&tcon->open_file_lock);
+		return cifs_get_writable_file(cinode, 0, ret_file);
+	}
+
+	spin_unlock(&tcon->open_file_lock);
+	return -ENOENT;
+}
+
 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
 {
 	struct address_space *mapping = page->mapping;
fs/cifs/smb2inode.c +68 -26

210 lines not on this page · raw message

Signature · 1 line
2.13.6
#1 lore raw

чт, 29 авг. 2019 г. в 15:26, Ronnie Sahlberg <[email protected]>:

Quoted 8 lines

rename() takes a path for old_file and in SMB2 we used to just create a compound for create(old_path)/rename/close().
If we already have a writable handle we can avoid the create() and close() altogether and just use the existing handle.

For this situation, as we avoid doing the create()
we also avoid triggering an oplock break for the existing handle.

Hi Ronnie,

Another SMB2_CREATE from the same client shouldn't cause a lease break if we ask for a lease with the same lease key. The lease key is associated with the inode, all opens that can cause a break should ask for the lease with the lease key.

Currently smb2_compound_op doesn't ask for a lease: see at the beginning of the function:

oplock = SMB2_OPLOCK_LEVEL_NONE

What we need to do is to change this to SMB2_OPLOCK_LEVEL_BATCH for operation that modifies the file like SET_EOF and SET_INFO. In this case the lower layer will map it to the appropriate lease state. Note that parms->fid->lease_key should be populated.

--
Best regards,
Pavel Shilovsky

#2 lore raw

Even so ... we shouldn't ask for open/query/close or open/setinfo/close if just a query-info or set-info (no open/close) is possible - it will make server's lives much easier ...

Quoted 31 lines · Pavel Shilovsky

чт, 29 авг. 2019 г. в 15:26, Ronnie Sahlberg <[email protected]>:

Quoted 8 lines

rename() takes a path for old_file and in SMB2 we used to just create a compound for create(old_path)/rename/close().
If we already have a writable handle we can avoid the create() and close() altogether and just use the existing handle.

For this situation, as we avoid doing the create()
we also avoid triggering an oplock break for the existing handle.

Hi Ronnie,

Another SMB2_CREATE from the same client shouldn't cause a lease break if we ask for a lease with the same lease key. The lease key is associated with the inode, all opens that can cause a break should ask for the lease with the lease key.

Currently smb2_compound_op doesn't ask for a lease: see at the beginning of the function:

oplock = SMB2_OPLOCK_LEVEL_NONE

What we need to do is to change this to SMB2_OPLOCK_LEVEL_BATCH for operation that modifies the file like SET_EOF and SET_INFO. In this case the lower layer will map it to the appropriate lease state. Note that parms->fid->lease_key should be populated.

--
Best regards,
Pavel Shilovsky

Signature · 3 lines
Thanks,

Steve