* [PATCH 0/1] cifs: do not leak EDEADLK to dgetents64
@ 2021-08-25 7:30 Ronnie Sahlberg
2021-08-25 7:30 ` [PATCH] cifs: Do not leak EDEADLK to dgetents64 for STATUS_USER_SESSION_DELETED Ronnie Sahlberg
0 siblings, 1 reply; 2+ messages in thread
From: Ronnie Sahlberg @ 2021-08-25 7:30 UTC (permalink / raw)
To: linux-cifs; +Cc: Steve French
Steve, List
Please find a patch that stops us from leaking EDEADLK (not enough credits)
to userspace when we do not have enough credits due to a pending reconnect.
This can be triggered for example if the server responds with
STATUS_USER_SESSION_DELETED during the Create part of the Create/QueryDir
that starts a directory scan.
Easiest way to reproduce this is patching up scrambla to inject this error
every 3 directory scans:
diff --git a/server/server.py b/server/server.py
index 7fd113b..47d0b7f 100644
--- a/server/server.py
+++ b/server/server.py
@@ -26,6 +26,7 @@ from smb2.filesystem_info import *
from smb2.dir_info import *
from smb2.ntlmssp import *
+
class File(object):
def __init__(self, path, flags, at, **kwargs):
@@ -81,6 +82,7 @@ class Server(object):
dialect = 0
def __init__(self, s, **kwargs):
+ self.errc = 0
self._s = s
self._sesid = 1
self._treeid = 1
@@ -348,6 +350,16 @@ class Server(object):
#
# Create/Open
#
+ #print('PDU', pdu)
+ if pdu['desired_access'] == 0x81:
+ print('YEAH')
+ self.errc = self.errc + 1
+ if self.errc == 3:
+ print('Generate error')
+ self.errc = 0
+ self._compound_error = Status.INVALID_PARAMETER
+ return (Status.USER_SESSION_DELETED,
+ ErrorResponse.encode({'error_data' : bytes(1)}))
if not hdr['tree_id'] in self.trees:
self._compound_error = Status.INVALID_PARAMETER
return (self._compound_error,
^ permalink raw reply related [flat|nested] 2+ messages in thread
* [PATCH] cifs: Do not leak EDEADLK to dgetents64 for STATUS_USER_SESSION_DELETED
2021-08-25 7:30 [PATCH 0/1] cifs: do not leak EDEADLK to dgetents64 Ronnie Sahlberg
@ 2021-08-25 7:30 ` Ronnie Sahlberg
0 siblings, 0 replies; 2+ messages in thread
From: Ronnie Sahlberg @ 2021-08-25 7:30 UTC (permalink / raw)
To: linux-cifs; +Cc: Steve French
RHBZ: 1994393
If we hit a STATUS_USER_SESSION_DELETED for the Create part in the
Create/QueryDirectory compound that starts a directory scan
we will leak EDEADLK back to userspace and surprise glibc and the application.
Pick this up cifs_readdir() and retry a small number of tries before we
return an error to userspace.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
---
fs/cifs/readdir.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/fs/cifs/readdir.c b/fs/cifs/readdir.c
index bfee176b901d..56e5d456366d 100644
--- a/fs/cifs/readdir.c
+++ b/fs/cifs/readdir.c
@@ -930,6 +930,7 @@ int cifs_readdir(struct file *file, struct dir_context *ctx)
unsigned int max_len;
const char *full_path;
void *page = alloc_dentry_path();
+ int retry_count = 0;
xid = get_xid();
@@ -944,8 +945,15 @@ int cifs_readdir(struct file *file, struct dir_context *ctx)
* '..'. Otherwise we won't be able to notify VFS in case of failure.
*/
if (file->private_data == NULL) {
+ again:
rc = initiate_cifs_search(xid, file, full_path);
- cifs_dbg(FYI, "initiate cifs search rc %d\n", rc);
+ if (rc == -EDEADLK && retry_count++ < 5) {
+ /*
+ * We don't have enough credits to start reading the
+ * directory so just try again.
+ */
+ goto again;
+ }
if (rc)
goto rddir2_exit;
}
--
2.30.2
^ permalink raw reply related [flat|nested] 2+ messages in thread
end of thread, other threads:[~2021-08-25 7:31 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-08-25 7:30 [PATCH 0/1] cifs: do not leak EDEADLK to dgetents64 Ronnie Sahlberg
2021-08-25 7:30 ` [PATCH] cifs: Do not leak EDEADLK to dgetents64 for STATUS_USER_SESSION_DELETED Ronnie Sahlberg
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).