| Message ID | 20190706191731.31150-1-aaptel@suse.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v1,4.20.y,stable] CIFS: fix deadlock in cached root handling | expand |
сб, 6 июл. 2019 г. в 12:17, Aurelien Aptel <aaptel@suse.com>: > > Prevent deadlock between open_shroot() and > cifs_mark_open_files_invalid() by releasing the lock before entering > SMB2_open, taking it again after and checking if we still need to use > the result. > > Link: https://lore.kernel.org/linux-cifs/684ed01c-cbca-2716-bc28-b0a59a0f8521@prodrive-technologies.com/T/#u > Fixes: 3d4ef9a15343 ("smb3: fix redundant opens on root") > Signed-off-by: Aurelien Aptel <aaptel@suse.com> > --- > > This patch applies on top of the 4.20 stable tree. > > Please review this carefuly, I have not tested it. > > XXX: do we need to call SMB2_close() when the work was already done > concurrently? if yes we need to do it outside the critical > section otherwise we hit the same issue again Yes, we do need to close a handle we just opened. Otherwise it will leak a handle on the server. > > fs/cifs/smb2ops.c | 22 ++++++++++++++++++++++ > 1 file changed, 22 insertions(+) > > diff --git a/fs/cifs/smb2ops.c b/fs/cifs/smb2ops.c > index aa71e620f3cd..55991e43d74f 100644 > --- a/fs/cifs/smb2ops.c > +++ b/fs/cifs/smb2ops.c > @@ -609,7 +609,27 @@ int open_shroot(unsigned int xid, struct cifs_tcon *tcon, struct cifs_fid *pfid) > oparams.fid = pfid; > oparams.reconnect = false; > > + /* > + * We do not hold the lock for the open because in case > + * SMB2_open needs to reconnect, it will end up calling > + * cifs_mark_open_files_invalid() which takes the lock again > + * thus causing a deadlock > + */ > + mutex_unlock(&tcon->crfid.fid_mutex); > rc = SMB2_open(xid, &oparams, &srch_path, &oplock, NULL, NULL, NULL); > + mutex_lock(&tcon->crfid.fid_mutex); > + > + /* > + * Now we need to check again as the cached root might have > + * been successfully re-opened from a concurrent process > + */ > + > + if (tcon->crfid.is_valid) { > + /* work was already done */ > + rc = 0; > + goto out; > + } > + > if (rc == 0) { > memcpy(tcon->crfid.fid, pfid, sizeof(struct cifs_fid)); > tcon->crfid.tcon = tcon; > @@ -617,6 +637,8 @@ int open_shroot(unsigned int xid, struct cifs_tcon *tcon, struct cifs_fid *pfid) > kref_init(&tcon->crfid.refcount); > kref_get(&tcon->crfid.refcount); > } > + > +out: > mutex_unlock(&tcon->crfid.fid_mutex); > return rc; > } > -- > 2.16.4 > -- Best regards, Pavel Shilovsky
diff --git a/fs/cifs/smb2ops.c b/fs/cifs/smb2ops.c index aa71e620f3cd..55991e43d74f 100644 --- a/fs/cifs/smb2ops.c +++ b/fs/cifs/smb2ops.c @@ -609,7 +609,27 @@ int open_shroot(unsigned int xid, struct cifs_tcon *tcon, struct cifs_fid *pfid) oparams.fid = pfid; oparams.reconnect = false; + /* + * We do not hold the lock for the open because in case + * SMB2_open needs to reconnect, it will end up calling + * cifs_mark_open_files_invalid() which takes the lock again + * thus causing a deadlock + */ + mutex_unlock(&tcon->crfid.fid_mutex); rc = SMB2_open(xid, &oparams, &srch_path, &oplock, NULL, NULL, NULL); + mutex_lock(&tcon->crfid.fid_mutex); + + /* + * Now we need to check again as the cached root might have + * been successfully re-opened from a concurrent process + */ + + if (tcon->crfid.is_valid) { + /* work was already done */ + rc = 0; + goto out; + } + if (rc == 0) { memcpy(tcon->crfid.fid, pfid, sizeof(struct cifs_fid)); tcon->crfid.tcon = tcon; @@ -617,6 +637,8 @@ int open_shroot(unsigned int xid, struct cifs_tcon *tcon, struct cifs_fid *pfid) kref_init(&tcon->crfid.refcount); kref_get(&tcon->crfid.refcount); } + +out: mutex_unlock(&tcon->crfid.fid_mutex); return rc; }
Prevent deadlock between open_shroot() and cifs_mark_open_files_invalid() by releasing the lock before entering SMB2_open, taking it again after and checking if we still need to use the result. Link: https://lore.kernel.org/linux-cifs/684ed01c-cbca-2716-bc28-b0a59a0f8521@prodrive-technologies.com/T/#u Fixes: 3d4ef9a15343 ("smb3: fix redundant opens on root") Signed-off-by: Aurelien Aptel <aaptel@suse.com> --- This patch applies on top of the 4.20 stable tree. Please review this carefuly, I have not tested it. XXX: do we need to call SMB2_close() when the work was already done concurrently? if yes we need to do it outside the critical section otherwise we hit the same issue again fs/cifs/smb2ops.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)