Skip to contentSkip to conversation

smb/server: fix some bugs in preparation for change notify support

PATCH Superseded ChenXiaoSong 3 participants 12 messages Jul 16 – Jul 17

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

ChenXiaoSong

author reviewer
#0 lore raw

v2 cover smb/server: fix some bugs in preparation for change notify support

From: ChenXiaoSong <[email protected]>

v1->v2:
- Update some coding style.
- Patch #03: use pr_err_ratelimited().

v1: https://lore.kernel.org/linux-cifs/[email protected]/

I have implemented the ksmbd change notify feature, please see the following GitHub link (some minor changes are also needed):
https://github.com/chenxiaosonggithub/linux/commits/ksmbd-change-notify/

However, to make this feature work properly, some pre-existing bugs need to be fixed first.

ChenXiaoSong (3):
smb/server: fix signing when a response uses more than one iov smb/server: cancel async requests when closing connection smb/server: stop new async work when closing connection

 fs/smb/server/connection.c | 20 ++++++++++
 fs/smb/server/smb2pdu.c    | 76 ++++++++++++++++++++++++++------------
 2 files changed, 73 insertions(+), 23 deletions(-)
Signature · 1 line
2.54.0
author reviewer
#1 lore raw

PATCH v2 1/3 smb/server: fix signing when a response uses more than one iov

From: ChenXiaoSong <[email protected]>

Some SMB responses keep their data in another buffer. The SMB header and the data are then in different iovs.

The old code only handled this for SMB2 READ. For other commands, it signed only the last iov. QUERY_INFO and CHANGE_NOTIFY can also use another iov for their data. Their SMB header was not signed, so Windows will client rejected the response.

Find the iov that starts with the current SMB header. Sign this iov and all iovs after it.

---
 fs/smb/server/smb2pdu.c | 52 ++++++++++++++++++++++++++++++-----------
 1 file changed, 38 insertions(+), 14 deletions(-)
fs/smb/server/smb2pdu.c +38 -14
index 295cca6cf3ae..966e61788422 100644
@@ -10624,6 +10624,40 @@ int smb2_check_sign_req(struct ksmbd_work *work)
 	return 1;
 }
 
+/**
+ * smb2_get_sign_rsp_iov() - get the iovecs used to sign a response
+ * @work: work that has the response iovecs
+ * @hdr: SMB2 header of the response
+ * @n_vec: set to the number of iovecs to sign
+ *
+ * Response data may be in another buffer. In this case, the response uses
+ * more than one iovec. Find the iovec that starts with @hdr. Sign this
+ * iovec and all iovecs after it.
+ *
+ * Return: The first iovec to sign.
+ */
+static struct kvec *smb2_get_sign_rsp_iov(struct ksmbd_work *work,
+					   struct smb2_hdr *hdr, int *n_vec)
+{
+	int i;
+
+	/*
+	 * iov[0] has the RFC1002 message length. It is not part of the SMB2
+	 * message, so do not sign it.
+	 */
+	for (i = 1; i <= work->iov_idx; i++) {
+		if (work->iov[i].iov_base == hdr) {
+			*n_vec = work->iov_idx - i + 1;
+			return &work->iov[i];
+		}
+	}
+
+	WARN_ON_ONCE(work->iov_idx < 1 ||
+		     work->iov[work->iov_idx].iov_base != hdr);
+	*n_vec = 1;
+	return &work->iov[work->iov_idx];
+}
+
 /**
  * smb2_set_sign_rsp() - handler for rsp packet sign processing
  * @work:   smb work containing notify command buffer
@@ -10634,18 +10668,13 @@ void smb2_set_sign_rsp(struct ksmbd_work *work)
 	struct smb2_hdr *hdr;
 	char signature[SMB2_HMACSHA256_SIZE];
 	struct kvec *iov;
-	int n_vec = 1;
+	int n_vec;
 
 	hdr = ksmbd_resp_buf_curr(work);
 	hdr->Flags |= SMB2_FLAGS_SIGNED;
 	memset(hdr->Signature, 0, SMB2_SIGNATURE_SIZE);
 
-	if (hdr->Command == SMB2_READ) {
-		iov = &work->iov[work->iov_idx - 1];
-		n_vec++;
-	} else {
-		iov = &work->iov[work->iov_idx];
-	}
+	iov = smb2_get_sign_rsp_iov(work, hdr, &n_vec);
 
 	ksmbd_sign_smb2_pdu(work->conn, work->sess->sess_key, iov, n_vec,
 			    signature);
@@ -10728,7 +10757,7 @@ void smb3_set_sign_rsp(struct ksmbd_work *work)
 	char signature[SMB2_CMACAES_SIZE];
 	struct kvec *iov;
 	u16 command = conn->ops->get_cmd_val(work);
-	int n_vec = 1;
+	int n_vec;
 	char *signing_key;
 
 	hdr = ksmbd_resp_buf_curr(work);
@@ -10750,12 +10779,7 @@ void smb3_set_sign_rsp(struct ksmbd_work *work)
 	hdr->Flags |= SMB2_FLAGS_SIGNED;
 	memset(hdr->Signature, 0, SMB2_SIGNATURE_SIZE);
 
-	if (hdr->Command == SMB2_READ) {
-		iov = &work->iov[work->iov_idx - 1];
-		n_vec++;
-	} else {
-		iov = &work->iov[work->iov_idx];
-	}
+	iov = smb2_get_sign_rsp_iov(work, hdr, &n_vec);
 
 	ksmbd_sign_smb3_pdu(conn, signing_key, iov, n_vec, signature);
 	memcpy(hdr->Signature, signature, SMB2_SIGNATURE_SIZE);
Signature · 1 line
2.54.0
#2 lore raw

Re: [PATCH v2 1/3] smb/server: fix signing when a response uses more than one iov

Quoted 10 lines · ChenXiaoSong

Some SMB responses keep their data in another buffer. The SMB header and the data are then in different iovs.

The old code only handled this for SMB2 READ. For other commands, it signed only the last iov. QUERY_INFO and CHANGE_NOTIFY can also use another iov for their data. Their SMB header was not signed, so Windows will client rejected the response.

Find the iov that starts with the current SMB header. Sign this iov and all iovs after it.

...

fs/smb/server/smb2pdu.c int smb2_check_sign_req(struct ksmbd_work *work) · line 10642

Quoted 5 lines
fs/smb/server/smb2pdu.c +4 -0
+static struct kvec *smb2_get_sign_rsp_iov(struct ksmbd_work *work,
+					   struct smb2_hdr *hdr, int *n_vec)
+{
 
+	int i;

This is not addressed, why?

fs/smb/server/smb2pdu.c int smb2_check_sign_req(struct ksmbd_work *work) · line 10656

Quoted 13 lines
fs/smb/server/smb2pdu.c +13 -0
+	/*
+	 * iov[0] has the RFC1002 message length. It is not part of the SMB2
+	 * message, so do not sign it.
+	 */
+	for (i = 1; i <= work->iov_idx; i++) {
+		if (work->iov[i].iov_base == hdr) {
+			*n_vec = work->iov_idx - i + 1;
+			return &work->iov[i];
+		}
+	}
+
+	WARN_ON_ONCE(work->iov_idx < 1 ||
+		     work->iov[work->iov_idx].iov_base != hdr);

The second part of this check is basically always true. What's the point?

Quoted 3 lines
fs/smb/server/smb2pdu.c +3 -0
+	*n_vec = 1;
+	return &work->iov[work->iov_idx];
+}
Signature · 2 lines
With Best Regards,
Andy Shevchenko
author reviewer
#3 lore raw

Re: [PATCH v2 1/3] smb/server: fix signing when a response uses more than one iov

The warning is triggered only when invalid arguments are passed.

Quoted 3 lines · Andy Shevchenko

fs/smb/server/smb2pdu.c int smb2_check_sign_req(struct ksmbd_work *work) · line 10656

Quoted 2 lines
fs/smb/server/smb2pdu.c +2 -0
+	WARN_ON_ONCE(work->iov_idx < 1 ||
+		     work->iov[work->iov_idx].iov_base != hdr);

The second part of this check is basically always true. What's the point?

Signature · 3 lines
ChenXiaoSong <[email protected]>
Chinese Homepage: https://chenxiaosong.com
English Homepage: https://chenxiaosong.com/en
#4 lore raw

Re: [PATCH v2 1/3] smb/server: fix signing when a response uses more than one iov

Quoted 1 line · ChenXiaoSong

The warning is triggered only when invalid arguments are passed.

Can you elaborate? The second part is the same check as it's done in the for-loop body. The first part checks if we ever go into that loop, so the logic here is either we have an invalid input (we skip the loop) or always true due to repetition of the check inside, meaning that the loop passed without finding the match.

So, what's the point?

Quoted 4 lines · Andy Shevchenko
Quoted 3 lines

fs/smb/server/smb2pdu.c int smb2_check_sign_req(struct ksmbd_work *work) · line 10656

Quoted 2 lines
fs/smb/server/smb2pdu.c +2 -0
+	WARN_ON_ONCE(work->iov_idx < 1 ||
+		     work->iov[work->iov_idx].iov_base != hdr);

The second part of this check is basically always true. What's the point?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
author reviewer
#5 lore raw

Re: [PATCH v2 1/3] smb/server: fix signing when a response uses more than one iov

The condition is always true. The explicit checks were intended to document the invalid state, but WARN_ON_ONCE(1) is clearer. I will update it. Thanks!

Quoted 7 lines · Andy Shevchenko

Can you elaborate? The second part is the same check as it's done in the for-loop body. The first part checks if we ever go into that loop, so the logic here is either we have an invalid input (we skip the loop) or always true due to repetition of the check inside, meaning that the loop passed without finding the match.

So, what's the point?

Signature · 3 lines
ChenXiaoSong <[email protected]>
Chinese Homepage: https://chenxiaosong.com
English Homepage: https://chenxiaosong.com/en
author reviewer
#6 lore raw

PATCH v2 2/3 smb/server: cancel async requests when closing connection

In reply to ChenXiaoSong

From: ChenXiaoSong <[email protected]>

An async request may still be waiting when a connection is closed. This can stop the connection from closing.

Cancel active async requests before waiting for them to finish.

---
 fs/smb/server/connection.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
fs/smb/server/connection.c +20 -0
index 2de1413ba64e..b31b72c07f83 100644
@@ -291,6 +291,25 @@ void ksmbd_conn_try_dequeue_request(struct ksmbd_work *work)
 	wake_up_all(&conn->req_running_q);
 }
 
+static void ksmbd_conn_cancel_async_requests(struct ksmbd_conn *conn)
+{
+	struct ksmbd_work *work;
+
+	ksmbd_debug(CONN, "Cancel pending async requests on releasing connection\n");
+	spin_lock(&conn->request_lock);
+	list_for_each_entry(work, &conn->async_requests, async_request_entry) {
+		if (work->state != KSMBD_WORK_ACTIVE)
+			continue;
+
+		ksmbd_debug(CONN, "Cancel async request id %d\n",
+			    work->async_id);
+		work->state = KSMBD_WORK_CANCELLED;
+		if (work->cancel_fn)
+			work->cancel_fn(work->cancel_argv);
+	}
+	spin_unlock(&conn->request_lock);
+}
+
 void ksmbd_conn_lock(struct ksmbd_conn *conn)
 {
 	mutex_lock(&conn->srv_mutex);
@@ -589,6 +608,7 @@ int ksmbd_conn_handler_loop(void *p)
 	}
 
 	ksmbd_conn_set_releasing(conn);
+	ksmbd_conn_cancel_async_requests(conn);
 	/* Wait till all reference dropped to the Server object*/
 	ksmbd_debug(CONN, "Wait for all pending requests(%d)\n", atomic_read(&conn->r_count));
 	wait_event(conn->r_count_q, atomic_read(&conn->r_count) == 0);
Signature · 1 line
2.54.0
maintainer
#7 lore raw

Re: [PATCH v2 2/3] smb/server: cancel async requests when closing connection

fs/smb/server/connection.c void ksmbd_conn_try_dequeue_request(struct ksmbd_work *work) · line 300

Quoted 7 lines
fs/smb/server/connection.c +7 -0
+static void ksmbd_conn_cancel_async_requests(struct ksmbd_conn *conn)
+{
+       struct ksmbd_work *work;
+
+       ksmbd_debug(CONN, "Cancel pending async requests on releasing connection\n");
+       spin_lock(&conn->request_lock);
+       list_for_each_entry(work, &conn->async_requests, async_request_entry) {

smb2_notify_cancel_fn() calls list_del_init() on async_request_entry. With list_for_each_entry(), the loop increment then follows the self-linked entry again, causing an infinite loop while holding request_lock ? So, should we use list_for_each_entry_safe() ?

Quoted 15 lines
fs/smb/server/connection.c +12 -0
+               if (work->state != KSMBD_WORK_ACTIVE)
+                       continue;
+
+               ksmbd_debug(CONN, "Cancel async request id %d\n",
+                           work->async_id);
+               work->state = KSMBD_WORK_CANCELLED;
+               if (work->cancel_fn)
+                       work->cancel_fn(work->cancel_argv);
+       }
+       spin_unlock(&conn->request_lock);
+}
+
 void ksmbd_conn_lock(struct ksmbd_conn *conn)
 {
        mutex_lock(&conn->srv_mutex);
author reviewer
#8 lore raw

Re: [PATCH v2 2/3] smb/server: cancel async requests when closing connection

Ok, list_for_each_entry_safe() would indeed be safer. I will use list_for_each_entry_safe() in the next version.

Quoted 5 lines · Namjae Jeon
Quoted 1 line

+ list_for_each_entry(work, &conn->async_requests, async_request_entry) {

smb2_notify_cancel_fn() calls list_del_init() on async_request_entry. With list_for_each_entry(), the loop increment then follows the self-linked entry again, causing an infinite loop while holding request_lock ? So, should we use list_for_each_entry_safe() ?

Signature · 3 lines
ChenXiaoSong <[email protected]>
Chinese Homepage: https://chenxiaosong.com
English Homepage: https://chenxiaosong.com/en
author reviewer
#9 lore raw

PATCH v2 3/3 smb/server: stop new async work when closing connection

In reply to ChenXiaoSong

From: ChenXiaoSong <[email protected]>

A new async request may be added while a connection is closing. The close code may miss this request and wait forever.

Check the connection state before adding the request. Return an error and release the async ID if the connection is closing.

---
 fs/smb/server/smb2pdu.c | 24 +++++++++++++++---------
 1 file changed, 15 insertions(+), 9 deletions(-)
fs/smb/server/smb2pdu.c +15 -9
index 966e61788422..b80f0bcf5fc1 100644
@@ -802,21 +802,27 @@ int setup_async_work(struct ksmbd_work *work, void (*fn)(void **), void **arg)
 		pr_err("Failed to alloc async message id\n");
 		return id;
 	}
-	work->asynchronous = true;
-	work->async_id = id;
 
-	ksmbd_debug(SMB,
-		    "Send interim Response to inform async request id : %d\n",
-		    work->async_id);
+	spin_lock(&conn->request_lock);
+	if (ksmbd_conn_exiting(conn) || ksmbd_conn_releasing(conn)) {
+		spin_unlock(&conn->request_lock);
+		pr_err_ratelimited("Failed to setup async work: connection is exiting\n");
+		ksmbd_release_id(&conn->async_ida, id);
+		return -ESHUTDOWN;
+	}
 
+	work->asynchronous = true;
+	work->async_id = id;
 	work->cancel_fn = fn;
 	work->cancel_argv = arg;
 
-	if (list_empty(&work->async_request_entry)) {
-		spin_lock(&conn->request_lock);
+	if (list_empty(&work->async_request_entry))
 		list_add_tail(&work->async_request_entry, &conn->async_requests);
-		spin_unlock(&conn->request_lock);
-	}
+	spin_unlock(&conn->request_lock);
+
+	ksmbd_debug(SMB,
+		    "Send interim Response to inform async request id : %d\n",
+		    work->async_id);
 
 	return 0;
 }
Signature · 1 line
2.54.0
maintainer
#10 lore raw

Re: [PATCH v2 3/3] smb/server: stop new async work when closing connection

Quoted 8 lines · ChenXiaoSong

From: ChenXiaoSong <[email protected]>

A new async request may be added while a connection is closing. The close code may miss this request and wait forever.

Check the connection state before adding the request. Return an error and release the async ID if the connection is closing.

setup_async_work() protects only work items registered through that helper. However, smb2_notify() later creates a synthetic in_work and adds it directly to conn->async_requests. Therefore, the connection-close path can finish scanning and cancelling the existing async requests before in_work is added. in_work is then registered after the connection enters the releasing state and is not cancelled. So, we need to protect this direct registration with the same connection-state check, or use a common helper for all async request registrations.
Thanks!

author reviewer
#11 lore raw

Re: [PATCH v2 3/3] smb/server: stop new async work when closing connection

Gaël will update his patch "ksmbd: defer CHANGE_NOTIFY completion instead of STATUS_NOT_IMPLEMENTED". Maybe the next version he won't use synthetic in_work.

Quoted 10 lines · Namjae Jeon

setup_async_work() protects only work items registered through that helper. However, smb2_notify() later creates a synthetic in_work and adds it directly to conn->async_requests. Therefore, the connection-close path can finish scanning and cancelling the existing async requests before in_work is added. in_work is then registered after the connection enters the releasing state and is not cancelled. So, we need to protect this direct registration with the same connection-state check, or use a common helper for all async request registrations.
Thanks!

Signature · 3 lines
ChenXiaoSong <[email protected]>
Chinese Homepage: https://chenxiaosong.com
English Homepage: https://chenxiaosong.com/en