Skip to contentSkip to conversation

kset: fix memory leak when kset_register() returns error

PATCH New Sasha Levin 1 participant 6 messages Dec 24, 2022

Subsystems (3)
lore
Share & tools
lore
author maintainer
#1 lore raw

PATCH 2/6 usb: gadget: aspeed: fix buffer overflow

In reply to Sasha Levin

From: Henry Tian <[email protected]>

[ Upstream commit 83045e19feae937c425248824d1dc0fc95583842 ]

In ast_vhub_epn_handle_ack() when the received data length exceeds the buffer, it does not check the case and just copies to req.buf and cause a buffer overflow, kernel oops on this case.

This issue could be reproduced on a BMC with an OS that enables the lan over USB:
1. In OS, enable the usb eth dev, verify it pings the BMC OK;
2. In OS, set the usb dev mtu to 2000. (Default is 1500);
3. In OS, ping the BMC with -s 2000 argument.

The BMC kernel will get oops with below logs:

    skbuff: skb_over_panic: text:8058e098 len:2048 put:2048 head:84c678a0 data:84c678c2 tail:0x84c680c2 end:0x84c67f00 dev:usb0
    ------------[ cut here ]------------
    kernel BUG at net/core/skbuff.c:113!
    Internal error: Oops - BUG: 0 [#1] ARM
    CPU: 0 PID: 0 Comm: swapper Not tainted 5.15.69-c9fb275-dirty-d1e579a #1
    Hardware name: Generic DT based system
    PC is at skb_panic+0x60/0x6c
    LR is at irq_work_queue+0x6c/0x94

Fix the issue by checking the length and set -EOVERFLOW.

Tested: Verify the BMC kernel does not get oops in the above case, and the usb ethernet gets RX packets errors instead.

---
 drivers/usb/gadget/udc/aspeed-vhub/core.c |  2 +-
 drivers/usb/gadget/udc/aspeed-vhub/epn.c  | 16 ++++++++++++----
 2 files changed, 13 insertions(+), 5 deletions(-)
drivers/usb/gadget/udc/aspeed-vhub/core.c +1 -1
index 902e61be4d64..7ce4dc132da3 100644
@@ -42,7 +42,7 @@ void ast_vhub_done(struct ast_vhub_ep *ep, struct ast_vhub_req *req,
 
 	list_del_init(&req->queue);
 
-	if (req->req.status == -EINPROGRESS)
+	if ((req->req.status == -EINPROGRESS) ||  (status == -EOVERFLOW))
 		req->req.status = status;
 
 	if (req->req.dma) {
drivers/usb/gadget/udc/aspeed-vhub/epn.c +12 -4
index 931f540a747e..32c381b3ed40 100644
@@ -89,6 +89,7 @@ static void ast_vhub_epn_handle_ack(struct ast_vhub_ep *ep)
 {
 	struct ast_vhub_req *req;
 	unsigned int len;
+	int status = 0;
 	u32 stat;
 
 	/* Read EP status */
@@ -124,9 +125,15 @@ static void ast_vhub_epn_handle_ack(struct ast_vhub_ep *ep)
 	len = VHUB_EP_DMA_TX_SIZE(stat);
 
 	/* If not using DMA, copy data out if needed */
-	if (!req->req.dma && !ep->epn.is_in && len)
-		memcpy(req->req.buf + req->req.actual, ep->buf, len);
-
+	if (!req->req.dma && !ep->epn.is_in && len) {
+		if (req->req.actual + len > req->req.length) {
+			req->last_desc = 1;
+			status = -EOVERFLOW;
+			goto done;
+		} else {
+			memcpy(req->req.buf + req->req.actual, ep->buf, len);
+		}
+	}
 	/* Adjust size */
 	req->req.actual += len;
 
@@ -134,9 +141,10 @@ static void ast_vhub_epn_handle_ack(struct ast_vhub_ep *ep)
 	if (len < ep->ep.maxpacket)
 		req->last_desc = 1;
 
+done:
 	/* That's it ? complete the request and pick a new one */
 	if (req->last_desc >= 0) {
-		ast_vhub_done(ep, req, 0);
+		ast_vhub_done(ep, req, status);
 		req = list_first_entry_or_null(&ep->queue, struct ast_vhub_req,
 					       queue);
 
Signature · 1 line
2.35.1

Sasha Levin

author maintainer
#0 lore raw

PATCH 1/6 kset: fix memory leak when kset_register() returns error

From: Yang Yingliang <[email protected]>

[ Upstream commit 1662cea4623f75d8251adf07370bbaa958f0355d ]

Inject fault while loading module, kset_register() may fail. If it fails, the kset.kobj.name allocated by kobject_set_name() which must be called before a call to kset_register() may be leaked, since refcount of kobj was set in kset_init().

To mitigate this, we free the name in kset_register() when an error is encountered, i.e. when kset_register() returns an error.

A kset may be embedded in a larger structure which may be dynamically allocated in callers, it needs to be freed in ktype.release() or error path in callers, in this case, we can not call kset_put() in kset_register(), or it will cause double free, so just call kfree_const() to free the name and set it to NULL to avoid accessing bad pointer in callers.

With this fix, the callers don't need care about freeing the name and may call kset_put() if kset_register() fails.

---
 lib/kobject.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
lib/kobject.c +8 -1
index 97d86dc17c42..1eb1230a2d28 100644
@@ -821,6 +821,9 @@ EXPORT_SYMBOL_GPL(kobj_sysfs_ops);
 /**
  * kset_register - initialize and add a kset.
  * @k: kset.
+ *
+ * NOTE: On error, the kset.kobj.name allocated by() kobj_set_name()
+ * is freed, it can not be used any more.
  */
 int kset_register(struct kset *k)
 {
@@ -831,8 +834,12 @@ int kset_register(struct kset *k)
 
 	kset_init(k);
 	err = kobject_add_internal(&k->kobj);
-	if (err)
+	if (err) {
+		kfree_const(k->kobj.name);
+		/* Set it to NULL to avoid accessing bad pointer in callers. */
+		k->kobj.name = NULL;
 		return err;
+	}
 	kobject_uevent(&k->kobj, KOBJ_ADD);
 	return 0;
 }
Signature · 1 line
2.35.1
author maintainer
#3 lore raw

PATCH 4/6 usb: gadget: f_ecm: Always set current gadget in ecm_bind()

From: Sascha Hauer <[email protected]>

[ Upstream commit d65e6b6e884a38360fc1cadf8ff31858151da57f ]

The gadget may change over bind/unbind cycles, so set it each time during bind, not only the first time. Without it we get a use-after-free with the following example:

cd /sys/kernel/config/usb_gadget/; mkdir -p mygadget; cd mygadget mkdir -p configs/c.1/strings/0x409
echo "C1:Composite Device" > configs/c.1/strings/0x409/configuration mkdir -p functions/ecm.usb0
ln -s functions/ecm.usb0 configs/c.1/
rmmod dummy_hcd
modprobe dummy_hcd

KASAN will complain shortly after the 'modprobe':

usb 2-1: New USB device found, idVendor=0000, idProduct=0000, bcdDevice= 6.01 usb 2-1: New USB device strings: Mfr=0, Product=0, SerialNumber=0
==================================================================

BUG: KASAN: use-after-free in gether_connect+0xb8/0x30c
Read of size 4 at addr cbef170c by task swapper/3/0

CPU: 3 PID: 0 Comm: swapper/3 Not tainted 6.1.0-rc3-00014-g41ff012f50cb-dirty #322
Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
 unwind_backtrace from show_stack+0x10/0x14
 show_stack from dump_stack_lvl+0x58/0x70
 dump_stack_lvl from print_report+0x134/0x4d4
 print_report from kasan_report+0x78/0x10c
 kasan_report from gether_connect+0xb8/0x30c
 gether_connect from ecm_set_alt+0x124/0x254
 ecm_set_alt from composite_setup+0xb98/0x2b18
 composite_setup from configfs_composite_setup+0x80/0x98
 configfs_composite_setup from dummy_timer+0x8f0/0x14a0 [dummy_hcd]
 ...
---
 drivers/usb/gadget/function/f_ecm.c | 22 +++++++++-------------
 1 file changed, 9 insertions(+), 13 deletions(-)
drivers/usb/gadget/function/f_ecm.c +9 -13
index ffe2486fce71..a7ab30e603e2 100644
@@ -685,7 +685,7 @@ ecm_bind(struct usb_configuration *c, struct usb_function *f)
 	struct usb_composite_dev *cdev = c->cdev;
 	struct f_ecm		*ecm = func_to_ecm(f);
 	struct usb_string	*us;
-	int			status;
+	int			status = 0;
 	struct usb_ep		*ep;
 
 	struct f_ecm_opts	*ecm_opts;
@@ -695,23 +695,19 @@ ecm_bind(struct usb_configuration *c, struct usb_function *f)
 
 	ecm_opts = container_of(f->fi, struct f_ecm_opts, func_inst);
 
-	/*
-	 * in drivers/usb/gadget/configfs.c:configfs_composite_bind()
-	 * configurations are bound in sequence with list_for_each_entry,
-	 * in each configuration its functions are bound in sequence
-	 * with list_for_each_entry, so we assume no race condition
-	 * with regard to ecm_opts->bound access
-	 */
+	mutex_lock(&ecm_opts->lock);
+
+	gether_set_gadget(ecm_opts->net, cdev->gadget);
+
 	if (!ecm_opts->bound) {
-		mutex_lock(&ecm_opts->lock);
-		gether_set_gadget(ecm_opts->net, cdev->gadget);
 		status = gether_register_netdev(ecm_opts->net);
-		mutex_unlock(&ecm_opts->lock);
-		if (status)
-			return status;
 		ecm_opts->bound = true;
 	}
 
+	mutex_unlock(&ecm_opts->lock);
+	if (status)
+		return status;
+
 	ecm_string_defs[1].s = ecm->ethaddr;
 
 	us = usb_gstrings_attach(cdev, ecm_strings,
Signature · 1 line
2.35.1
author maintainer
#2 lore raw

PATCH 3/6 usb: gadget: u_ether: Do not make UDC parent of the net device

In reply to Sasha Levin

From: Sascha Hauer <[email protected]>

[ Upstream commit 321b59870f850a10dbb211ecd2bd87b41497ea6f ]

The UDC is not a suitable parent of the net device as the UDC can change or vanish during the lifecycle of the ethernet gadget. This can be illustrated with the following:

mkdir -p /sys/kernel/config/usb_gadget/mygadget
cd /sys/kernel/config/usb_gadget/mygadget
mkdir -p configs/c.1/strings/0x409
echo "C1:Composite Device" > configs/c.1/strings/0x409/configuration mkdir -p functions/ecm.usb0
ln -s functions/ecm.usb0 configs/c.1/
echo "dummy_udc.0" > UDC
rmmod dummy_hcd

The 'rmmod' removes the UDC from the just created gadget, leaving the still existing net device with a no longer existing parent.

Accessing the ethernet device with commands like:

ip --details link show usb0

will result in a KASAN splat:

==================================================================

BUG: KASAN: use-after-free in if_nlmsg_size+0x3e8/0x528
Read of size 4 at addr c5c84754 by task ip/357

CPU: 3 PID: 357 Comm: ip Not tainted 6.1.0-rc3-00013-gd14953726b24-dirty #324
Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
 unwind_backtrace from show_stack+0x10/0x14
 show_stack from dump_stack_lvl+0x58/0x70
 dump_stack_lvl from print_report+0x134/0x4d4
 print_report from kasan_report+0x78/0x10c
 kasan_report from if_nlmsg_size+0x3e8/0x528
 if_nlmsg_size from rtnl_getlink+0x2b4/0x4d0
 rtnl_getlink from rtnetlink_rcv_msg+0x1f4/0x674
 rtnetlink_rcv_msg from netlink_rcv_skb+0xb4/0x1f8
 netlink_rcv_skb from netlink_unicast+0x294/0x478
 netlink_unicast from netlink_sendmsg+0x328/0x640
 netlink_sendmsg from ____sys_sendmsg+0x2a4/0x3b4
 ____sys_sendmsg from ___sys_sendmsg+0xc8/0x12c
 ___sys_sendmsg from sys_sendmsg+0xa0/0x120
 sys_sendmsg from ret_fast_syscall+0x0/0x1c

Solve this by not setting the parent of the ethernet device.

---
 drivers/usb/gadget/function/u_ether.c | 4 ----
 1 file changed, 4 deletions(-)
drivers/usb/gadget/function/u_ether.c +0 -4
index 2fe91f120bb1..c4196c32c0be 100644
@@ -795,7 +795,6 @@ struct eth_dev *gether_setup_name(struct usb_gadget *g,
 	net->max_mtu = GETHER_MAX_MTU_SIZE;
 
 	dev->gadget = g;
-	SET_NETDEV_DEV(net, &g->dev);
 	SET_NETDEV_DEVTYPE(net, &gadget_type);
 
 	status = register_netdev(net);
@@ -869,8 +868,6 @@ int gether_register_netdev(struct net_device *net)
 	struct usb_gadget *g;
 	int status;
 
-	if (!net->dev.parent)
-		return -EINVAL;
 	dev = netdev_priv(net);
 	g = dev->gadget;
 
@@ -901,7 +898,6 @@ void gether_set_gadget(struct net_device *net, struct usb_gadget *g)
 
 	dev = netdev_priv(net);
 	dev->gadget = g;
-	SET_NETDEV_DEV(net, &g->dev);
 }
 EXPORT_SYMBOL_GPL(gether_set_gadget);
 
Signature · 1 line
2.35.1
author maintainer
#4 lore raw

PATCH 5/6 usb/usbip: Fix v_recv_cmd_submit() to use PIPE_BULK define

In reply to Sasha Levin

From: Shuah Khan <[email protected]>

[ Upstream commit dd65a243a915ca319ed5fee9161a168c836fa2f2 ]

Fix v_recv_cmd_submit() to use PIPE_BULK define instead of hard coded values. This also fixes the following signed integer overflow error reported by cppcheck. This is not an issue since pipe is unsigned int. However, this change improves the code to use proper define.

drivers/usb/usbip/vudc_rx.c:152:26: error: Signed integer overflow for expression '3<<30'. [integerOverflow]
urb_p->urb->pipe &= ~(3 << 30);

In addition, add a build time check for PIPE_BULK != 3 as the code path depends on PIPE_BULK = 3.

---
 drivers/usb/usbip/vudc_rx.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/usb/usbip/vudc_rx.c +3 -1
index 1e8a23d92cb4..4ccff3efee4e 100644
@@ -149,7 +149,9 @@ static int v_recv_cmd_submit(struct vudc *udc,
 	urb_p->urb->status = -EINPROGRESS;
 
 	/* FIXME: more pipe setup to please usbip_common */
-	urb_p->urb->pipe &= ~(3 << 30);
+	BUILD_BUG_ON_MSG(PIPE_BULK != 3, "PIPE_* doesn't range from 0 to 3");
+
+	urb_p->urb->pipe &= ~(PIPE_BULK << 30);
 	switch (urb_p->ep->type) {
 	case USB_ENDPOINT_XFER_BULK:
 		urb_p->urb->pipe |= (PIPE_BULK << 30);
Signature · 1 line
2.35.1
author maintainer
#5 lore raw

PATCH 6/6 ACPICA: Fix operand resolution

In reply to Sasha Levin

From: Amadeusz Sławiński <[email protected]>

[ Upstream commit 7dfb216eda99bbfc2a8c3b03d2eec63314f52b3c ]

In our tests we get UBSAN warning coming from ACPI parser. This is caused by trying to resolve operands when there is none.

[    0.000000] Linux version 5.15.0-rc3chromeavsrel1.0.184+ (root@...) (gcc (Ubuntu 10.3.0-1ubuntu1~20.04) 10.3.0, GNU ld (GNU Binutils for Ubuntu) 2.34) #1 SMP PREEMPT Sat Oct 16 00:08:27 UTC 2021
...
[ 14.719508] ================================================================================
[ 14.719551] UBSAN: array-index-out-of-bounds in /.../linux/drivers/acpi/acpica/dswexec.c:401:12
[ 14.719594] index -1 is out of range for type 'acpi_operand_object *[9]'
[ 14.719621] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.15.0-rc3chromeavsrel1.0.184+ #1
[ 14.719657] Hardware name: Intel Corp. Geminilake/GLK RVP2 LP4SD (07), BIOS GELKRVPA.X64.0214.B50.2009111159 09/11/2020
[ 14.719694] Call Trace:
[ 14.719712] dump_stack_lvl+0x38/0x49
[ 14.719749] dump_stack+0x10/0x12
[ 14.719775] ubsan_epilogue+0x9/0x45
[ 14.719801] __ubsan_handle_out_of_bounds.cold+0x44/0x49
[ 14.719835] acpi_ds_exec_end_op+0x1d7/0x6b5
[ 14.719870] acpi_ps_parse_loop+0x942/0xb34

...

Problem happens because WalkState->NumOperands is 0 and it is used when trying to access into operands table. Actual code is:
WalkState->Operands [WalkState->NumOperands -1]
which causes out of bound access. Improve the check before above access to check if ACPI opcode should have any arguments (operands) at all.

---
 drivers/acpi/acpica/dswexec.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/acpi/acpica/dswexec.c +4 -2
index 1504b93cc5f4..93dfca60701f 100644
@@ -382,9 +382,11 @@ acpi_status acpi_ds_exec_end_op(struct acpi_walk_state *walk_state)
 
 		/*
 		 * All opcodes require operand resolution, with the only exceptions
-		 * being the object_type and size_of operators.
+		 * being the object_type and size_of operators as well as opcodes that
+		 * take no arguments.
 		 */
-		if (!(walk_state->op_info->flags & AML_NO_OPERAND_RESOLVE)) {
+		if (!(walk_state->op_info->flags & AML_NO_OPERAND_RESOLVE) &&
+		    (walk_state->op_info->flags & AML_HAS_ARGS)) {
 
 			/* Resolve all operands */
 
Signature · 1 line
2.35.1