Skip to contentSkip to conversation

tty: Allow TIOCSTI to be disabled

PATCH In review Kees Cook 8 participants 21 messages Oct 22, 2022 – Jul 3, 2023 TTY layer and serial drivers

lore
Share & tools
lore

Kees Cook

author
#0 lore raw

v3 cover tty: Allow TIOCSTI to be disabled

Hi,

v3: make sysctl a bool (Jiri)
v2: https://lore.kernel.org/lkml/[email protected]
v1: https://lore.kernel.org/lkml/[email protected]/

This is so very long over-due. We just need to do this and put an end to this legacy operation.

Repeating the commit log from patch 2:

TIOCSTI continues its long history of being used in privilege escalation attacks[1]. Prior attempts to provide a mechanism to disable this have devolved into discussions around creating full-blown LSMs to provide arbitrary ioctl filtering, which is hugely over-engineered -- only TIOCSTI is being used this way. 3 years ago OpenBSD entirely removed TIOCSTI[2], Android has had it filtered for longer[3], and the tools that had historically used TIOCSTI either do not need it, are not commonly built with it, or have had its use removed.

Provide a simple CONFIG and global sysctl to disable this for the system builders who have wanted this functionality for literally decades now, much like the ldisc_autoload CONFIG and sysctl.

[1] https://lore.kernel.org/linux-hardening/Y0m9l52AKmw6Yxi1@hostpad
[2] https://undeadly.org/cgi?action=article;sid=20170701132619
[3] https://lore.kernel.org/lkml/CAFJ0LnFGRuEEn1tCLhoki8ZyWrKfktbF+rwwN7WzyC_kBFoQVA@mail.gmail.com/

Thanks,

-Kees

Kees Cook (2):
tty: Move sysctl setup into "core" tty logic
tty: Allow TIOCSTI to be disabled

 drivers/tty/Kconfig     | 19 +++++++++++++++++
 drivers/tty/tty.h       |  2 +-
 drivers/tty/tty_io.c    | 45 +++++++++++++++++++++++++++++++++++++++--
 drivers/tty/tty_ldisc.c | 38 +---------------------------------
 4 files changed, 64 insertions(+), 40 deletions(-)
Signature · 1 line
2.34.1
author
#1 lore raw

PATCH v3 1/2 tty: Move sysctl setup into "core" tty logic

In preparation for adding another sysctl to the tty subsystem, move the tty setup code into the "core" tty code, which contains tty_init() itself.

Cc Greg Kroah-Hartman <[email protected]>
Cc Jiri Slaby <[email protected]>
---
 drivers/tty/tty.h       |  2 +-
 drivers/tty/tty_io.c    | 34 ++++++++++++++++++++++++++++++++--
 drivers/tty/tty_ldisc.c | 38 +-------------------------------------
 3 files changed, 34 insertions(+), 40 deletions(-)
drivers/tty/tty.h +1 -1
index 1c08c9b67b16..f45cd683c02e 100644
@@ -93,7 +93,7 @@ void tty_ldisc_release(struct tty_struct *tty);
 int __must_check tty_ldisc_init(struct tty_struct *tty);
 void tty_ldisc_deinit(struct tty_struct *tty);
 
-void tty_sysctl_init(void);
+extern int tty_ldisc_autoload;
 
 /* tty_audit.c */
 #ifdef CONFIG_AUDIT
drivers/tty/tty_io.c +32 -2
index de06c3c2ff70..fe77a3d41326 100644
@@ -3572,13 +3572,44 @@ void console_sysfs_notify(void)
 		sysfs_notify(&consdev->kobj, NULL, "active");
 }
 
+static struct ctl_table tty_table[] = {
+	{
+		.procname	= "ldisc_autoload",
+		.data		= &tty_ldisc_autoload,
+		.maxlen		= sizeof(tty_ldisc_autoload),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE,
+	},
+	{ }
+};
+
+static struct ctl_table tty_dir_table[] = {
+	{
+		.procname	= "tty",
+		.mode		= 0555,
+		.child		= tty_table,
+	},
+	{ }
+};
+
+static struct ctl_table tty_root_table[] = {
+	{
+		.procname	= "dev",
+		.mode		= 0555,
+		.child		= tty_dir_table,
+	},
+	{ }
+};
+
 /*
  * Ok, now we can initialize the rest of the tty devices and can count
  * on memory allocations, interrupts etc..
  */
 int __init tty_init(void)
 {
-	tty_sysctl_init();
+	register_sysctl_table(tty_root_table);
 	cdev_init(&tty_cdev, &tty_fops);
 	if (cdev_add(&tty_cdev, MKDEV(TTYAUX_MAJOR, 0), 1) ||
 	    register_chrdev_region(MKDEV(TTYAUX_MAJOR, 0), 1, "/dev/tty") < 0)
@@ -3600,4 +3631,3 @@ int __init tty_init(void)
 #endif
 	return 0;
 }
-
drivers/tty/tty_ldisc.c +1 -37
index 776d8a62f77c..e758f44729e7 100644
@@ -117,7 +117,7 @@ static void put_ldops(struct tty_ldisc_ops *ldops)
 	raw_spin_unlock_irqrestore(&tty_ldiscs_lock, flags);
 }
 
-static int tty_ldisc_autoload = IS_BUILTIN(CONFIG_LDISC_AUTOLOAD);
+int tty_ldisc_autoload = IS_BUILTIN(CONFIG_LDISC_AUTOLOAD);
 
 /**
  * tty_ldisc_get	-	take a reference to an ldisc
@@ -817,39 +817,3 @@ void tty_ldisc_deinit(struct tty_struct *tty)
 		tty_ldisc_put(tty->ldisc);
 	tty->ldisc = NULL;
 }
-
-static struct ctl_table tty_table[] = {
-	{
-		.procname	= "ldisc_autoload",
-		.data		= &tty_ldisc_autoload,
-		.maxlen		= sizeof(tty_ldisc_autoload),
-		.mode		= 0644,
-		.proc_handler	= proc_dointvec,
-		.extra1		= SYSCTL_ZERO,
-		.extra2		= SYSCTL_ONE,
-	},
-	{ }
-};
-
-static struct ctl_table tty_dir_table[] = {
-	{
-		.procname	= "tty",
-		.mode		= 0555,
-		.child		= tty_table,
-	},
-	{ }
-};
-
-static struct ctl_table tty_root_table[] = {
-	{
-		.procname	= "dev",
-		.mode		= 0555,
-		.child		= tty_dir_table,
-	},
-	{ }
-};
-
-void tty_sysctl_init(void)
-{
-	register_sysctl_table(tty_root_table);
-}
Signature · 1 line
2.34.1
author
#2 lore raw

PATCH v3 2/2 tty: Allow TIOCSTI to be disabled

In reply to Kees Cook

TIOCSTI continues its long history of being used in privilege escalation attacks[1]. Prior attempts to provide a mechanism to disable this have devolved into discussions around creating full-blown LSMs to provide arbitrary ioctl filtering, which is hugely over-engineered -- only TIOCSTI is being used this way. 3 years ago OpenBSD entirely removed TIOCSTI[2], Android has had it filtered for longer[3], and the tools that had historically used TIOCSTI either do not need it, are not commonly built with it, or have had its use removed.

Provide a simple CONFIG and global sysctl to disable this for the system builders who have wanted this functionality for literally decades now, much like the ldisc_autoload CONFIG and sysctl.

[1] https://lore.kernel.org/linux-hardening/Y0m9l52AKmw6Yxi1@hostpad
[2] https://undeadly.org/cgi?action=article;sid=20170701132619
[3] https://lore.kernel.org/lkml/CAFJ0LnFGRuEEn1tCLhoki8ZyWrKfktbF+rwwN7WzyC_kBFoQVA@mail.gmail.com/

Cc Greg Kroah-Hartman <[email protected]>
Cc Jiri Slaby <[email protected]>
Cc Simon Brand <[email protected]>
---
 drivers/tty/Kconfig  | 19 +++++++++++++++++++
 drivers/tty/tty_io.c | 11 +++++++++++
 2 files changed, 30 insertions(+)
drivers/tty/Kconfig +19 -0
index cc30ff93e2e4..d35fc068da74 100644
@@ -149,6 +149,25 @@ config LEGACY_PTY_COUNT
 	  When not in use, each legacy PTY occupies 12 bytes on 32-bit
 	  architectures and 24 bytes on 64-bit architectures.
 
+config LEGACY_TIOCSTI
+	bool "Allow legacy TIOCSTI usage"
+	default y
+	help
+	  Historically the kernel has allowed TIOCSTI, which will push
+	  characters into a controlling TTY. This continues to be used
+	  as a malicious privilege escalation mechanism, and provides no
+	  meaningful real-world utility any more. Its use is considered
+	  a dangerous legacy operation, and can be disabled on most
+	  systems.
+
+	  Say 'Y here only if you have confirmed that your system's
+	  userspace depends on this functionality to continue operating
+	  normally.
+
+	  This functionality can be changed at runtime with the
+	  dev.tty.legacy_tiocsti sysctl. This configuration option sets
+	  the default value of the sysctl.
+
 config LDISC_AUTOLOAD
 	bool "Automatically load TTY Line Disciplines"
 	default y
drivers/tty/tty_io.c +11 -0
index fe77a3d41326..a6a16cf986b7 100644
@@ -2268,11 +2268,15 @@ static int tty_fasync(int fd, struct file *filp, int on)
  *  * Called functions take tty_ldiscs_lock
  *  * current->signal->tty check is safe without locks
  */
+static bool tty_legacy_tiocsti __read_mostly = IS_ENABLED(CONFIG_LEGACY_TIOCSTI);
 static int tiocsti(struct tty_struct *tty, char __user *p)
 {
 	char ch, mbz = 0;
 	struct tty_ldisc *ld;
 
+	if (!tty_legacy_tiocsti)
+		return -EIO;
+
 	if ((current->signal->tty != tty) && !capable(CAP_SYS_ADMIN))
 		return -EPERM;
 	if (get_user(ch, p))
@@ -3573,6 +3577,13 @@ void console_sysfs_notify(void)
 }
 
 static struct ctl_table tty_table[] = {
+	{
+		.procname	= "legacy_tiocsti",
+		.data		= &tty_legacy_tiocsti,
+		.maxlen		= sizeof(tty_legacy_tiocsti),
+		.mode		= 0644,
+		.proc_handler	= proc_dobool,
+	},
 	{
 		.procname	= "ldisc_autoload",
 		.data		= &tty_ldisc_autoload,
Signature · 1 line
2.34.1
#3 lore raw

Hi Kees,

Quoted 13 lines · Kees Cook

TIOCSTI continues its long history of being used in privilege escalation attacks[1]. Prior attempts to provide a mechanism to disable this have devolved into discussions around creating full-blown LSMs to provide arbitrary ioctl filtering, which is hugely over-engineered -- only TIOCSTI is being used this way. 3 years ago OpenBSD entirely removed TIOCSTI[2], Android has had it filtered for longer[3], and the tools that had historically used TIOCSTI either do not need it, are not commonly built with it, or have had its use removed.

Provide a simple CONFIG and global sysctl to disable this for the system builders who have wanted this functionality for literally decades now, much like the ldisc_autoload CONFIG and sysctl.

Quoted 8 lines

[1] https://lore.kernel.org/linux-hardening/Y0m9l52AKmw6Yxi1@hostpad
[2] https://undeadly.org/cgi?action=article;sid=20170701132619
[3] https://lore.kernel.org/lkml/CAFJ0LnFGRuEEn1tCLhoki8ZyWrKfktbF+rwwN7WzyC_kBFoQVA@mail.gmail.com/

Cc Greg Kroah-Hartman <[email protected]>
Cc Jiri Slaby <[email protected]>
Cc Simon Brand <[email protected]>

Thanks for your patch, which is now commit 83efeeeb3d04b22a ("tty:
Allow TIOCSTI to be disabled") in tty/tty-next.

drivers/tty/Kconfig config LEGACY_PTY_COUNT · line 154

Quoted 9 lines
drivers/tty/Kconfig +3 -0
@@ -149,6 +149,25 @@ config LEGACY_PTY_COUNT
          When not in use, each legacy PTY occupies 12 bytes on 32-bit
          architectures and 24 bytes on 64-bit architectures.
 
+config LEGACY_TIOCSTI
+       bool "Allow legacy TIOCSTI usage"
+       default y

Obviously this should either default to n, ...

drivers/tty/Kconfig config LEGACY_PTY_COUNT · line 165

Quoted 11 lines
drivers/tty/Kconfig +11 -0
+       help
+         Historically the kernel has allowed TIOCSTI, which will push
+         characters into a controlling TTY. This continues to be used
+         as a malicious privilege escalation mechanism, and provides no
+         meaningful real-world utility any more. Its use is considered
+         a dangerous legacy operation, and can be disabled on most
+         systems.
+
+         Say 'Y here only if you have confirmed that your system's
+         userspace depends on this functionality to continue operating
+         normally.

... or the help text should be made less scary.

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

#4 lore raw
In reply to Kees Cook

Hello,

Kees Cook, le sam. 22 oct. 2022 11:29:49 -0700, a ecrit:

Quoted 8 lines

TIOCSTI continues its long history of being used in privilege escalation attacks[1]. Prior attempts to provide a mechanism to disable this have devolved into discussions around creating full-blown LSMs to provide arbitrary ioctl filtering, which is hugely over-engineered -- only TIOCSTI is being used this way. 3 years ago OpenBSD entirely removed TIOCSTI[2], Android has had it filtered for longer[3], and the tools that had historically used TIOCSTI either do not need it, are not commonly built with it, or have had its use removed.

No. The Brltty screen reader entirely relies on TIOCSTI to be able to support input from various Braille devices. Please make sure to keep TIOCSTI enabled by default, otherwise some people would just completely lose their usual way of simply typing on Linux.

Samuel

Quoted 23 lines

Provide a simple CONFIG and global sysctl to disable this for the system builders who have wanted this functionality for literally decades now, much like the ldisc_autoload CONFIG and sysctl.

[1] https://lore.kernel.org/linux-hardening/Y0m9l52AKmw6Yxi1@hostpad
[2] https://undeadly.org/cgi?action=article;sid=20170701132619
[3] https://lore.kernel.org/lkml/CAFJ0LnFGRuEEn1tCLhoki8ZyWrKfktbF+rwwN7WzyC_kBFoQVA@mail.gmail.com/

Cc Greg Kroah-Hartman <[email protected]>
Cc Jiri Slaby <[email protected]>
Cc Simon Brand <[email protected]>
---
 drivers/tty/Kconfig  | 19 +++++++++++++++++++
 drivers/tty/tty_io.c | 11 +++++++++++
 2 files changed, 30 insertions(+)
drivers/tty/Kconfig +0 -0
index cc30ff93e2e4..d35fc068da74 100644
@@ -149,6 +149,25 @@ config LEGACY_PTY_COUNT
 	  When not in use, each legacy PTY occupies 12 bytes on 32-bit

drivers/tty/Kconfig config LEGACY_PTY_COUNT · line 159

Quoted 10 lines
drivers/tty/Kconfig +8 -0
 	  architectures and 24 bytes on 64-bit architectures.
 
+config LEGACY_TIOCSTI
+	bool "Allow legacy TIOCSTI usage"
+	default y
+	help
+	  Historically the kernel has allowed TIOCSTI, which will push
+	  characters into a controlling TTY. This continues to be used
+	  as a malicious privilege escalation mechanism, and provides no
+	  meaningful real-world utility any more.

Yes it does.

Quoted 52 lines
drivers/tty/Kconfig +12 -0
+       Its use is considered
+	  a dangerous legacy operation, and can be disabled on most
+	  systems.
+
+	  Say 'Y here only if you have confirmed that your system's
+	  userspace depends on this functionality to continue operating
+	  normally.
+
+	  This functionality can be changed at runtime with the
+	  dev.tty.legacy_tiocsti sysctl. This configuration option sets
+	  the default value of the sysctl.
+
 config LDISC_AUTOLOAD
 	bool "Automatically load TTY Line Disciplines"
 	default y
drivers/tty/tty_io.c +11 -0
index fe77a3d41326..a6a16cf986b7 100644
@@ -2268,11 +2268,15 @@ static int tty_fasync(int fd, struct file *filp, int on)
  *  * Called functions take tty_ldiscs_lock
  *  * current->signal->tty check is safe without locks
  */
+static bool tty_legacy_tiocsti __read_mostly = IS_ENABLED(CONFIG_LEGACY_TIOCSTI);
 static int tiocsti(struct tty_struct *tty, char __user *p)
 {
 	char ch, mbz = 0;
 	struct tty_ldisc *ld;
 
+	if (!tty_legacy_tiocsti)
+		return -EIO;
+
 	if ((current->signal->tty != tty) && !capable(CAP_SYS_ADMIN))
 		return -EPERM;
 	if (get_user(ch, p))
@@ -3573,6 +3577,13 @@ void console_sysfs_notify(void)
 }
 
 static struct ctl_table tty_table[] = {
+	{
+		.procname	= "legacy_tiocsti",
+		.data		= &tty_legacy_tiocsti,
+		.maxlen		= sizeof(tty_legacy_tiocsti),
+		.mode		= 0644,
+		.proc_handler	= proc_dobool,
+	},
 	{
 		.procname	= "ldisc_autoload",
 		.data		= &tty_ldisc_autoload,
Signature · 1 line
2.34.1
Signature · 4 lines
Samuel
---
Pour une évaluation indépendante, transparente et rigoureuse !
Je soutiens la Commission d'Évaluation de l'Inria.
#5 lore raw

Samuel Thibault, le mer. 28 déc. 2022 00:40:00 +0100, a ecrit:

Quoted 12 lines

Hello,

Kees Cook, le sam. 22 oct. 2022 11:29:49 -0700, a ecrit:

Quoted 8 lines

TIOCSTI continues its long history of being used in privilege escalation attacks[1]. Prior attempts to provide a mechanism to disable this have devolved into discussions around creating full-blown LSMs to provide arbitrary ioctl filtering, which is hugely over-engineered -- only TIOCSTI is being used this way. 3 years ago OpenBSD entirely removed TIOCSTI[2], Android has had it filtered for longer[3], and the tools that had historically used TIOCSTI either do not need it, are not commonly built with it, or have had its use removed.

Quoted 2 lines

No. The Brltty screen reader entirely relies on TIOCSTI to be able to support input from various Braille devices.

(it only needs support for it on the linux console itself, nowhere else)

Quoted 100 lines

Please make sure to keep
TIOCSTI enabled by default, otherwise some people would just completely lose their usual way of simply typing on Linux.

Samuel

Quoted 23 lines

Provide a simple CONFIG and global sysctl to disable this for the system builders who have wanted this functionality for literally decades now, much like the ldisc_autoload CONFIG and sysctl.

[1] https://lore.kernel.org/linux-hardening/Y0m9l52AKmw6Yxi1@hostpad
[2] https://undeadly.org/cgi?action=article;sid=20170701132619
[3] https://lore.kernel.org/lkml/CAFJ0LnFGRuEEn1tCLhoki8ZyWrKfktbF+rwwN7WzyC_kBFoQVA@mail.gmail.com/

Cc Greg Kroah-Hartman <[email protected]>
Cc Jiri Slaby <[email protected]>
Cc Simon Brand <[email protected]>
---
 drivers/tty/Kconfig  | 19 +++++++++++++++++++
 drivers/tty/tty_io.c | 11 +++++++++++
 2 files changed, 30 insertions(+)
drivers/tty/Kconfig +0 -0
index cc30ff93e2e4..d35fc068da74 100644
@@ -149,6 +149,25 @@ config LEGACY_PTY_COUNT
 	  When not in use, each legacy PTY occupies 12 bytes on 32-bit

drivers/tty/Kconfig config LEGACY_PTY_COUNT · line 159

Quoted 10 lines
drivers/tty/Kconfig +8 -0
 	  architectures and 24 bytes on 64-bit architectures.
 
+config LEGACY_TIOCSTI
+	bool "Allow legacy TIOCSTI usage"
+	default y
+	help
+	  Historically the kernel has allowed TIOCSTI, which will push
+	  characters into a controlling TTY. This continues to be used
+	  as a malicious privilege escalation mechanism, and provides no
+	  meaningful real-world utility any more.

Yes it does.

Quoted 52 lines
drivers/tty/Kconfig +12 -0
+       Its use is considered
+	  a dangerous legacy operation, and can be disabled on most
+	  systems.
+
+	  Say 'Y here only if you have confirmed that your system's
+	  userspace depends on this functionality to continue operating
+	  normally.
+
+	  This functionality can be changed at runtime with the
+	  dev.tty.legacy_tiocsti sysctl. This configuration option sets
+	  the default value of the sysctl.
+
 config LDISC_AUTOLOAD
 	bool "Automatically load TTY Line Disciplines"
 	default y
drivers/tty/tty_io.c +11 -0
index fe77a3d41326..a6a16cf986b7 100644
@@ -2268,11 +2268,15 @@ static int tty_fasync(int fd, struct file *filp, int on)
  *  * Called functions take tty_ldiscs_lock
  *  * current->signal->tty check is safe without locks
  */
+static bool tty_legacy_tiocsti __read_mostly = IS_ENABLED(CONFIG_LEGACY_TIOCSTI);
 static int tiocsti(struct tty_struct *tty, char __user *p)
 {
 	char ch, mbz = 0;
 	struct tty_ldisc *ld;
 
+	if (!tty_legacy_tiocsti)
+		return -EIO;
+
 	if ((current->signal->tty != tty) && !capable(CAP_SYS_ADMIN))
 		return -EPERM;
 	if (get_user(ch, p))
@@ -3573,6 +3577,13 @@ void console_sysfs_notify(void)
 }
 
 static struct ctl_table tty_table[] = {
+	{
+		.procname	= "legacy_tiocsti",
+		.data		= &tty_legacy_tiocsti,
+		.maxlen		= sizeof(tty_legacy_tiocsti),
+		.mode		= 0644,
+		.proc_handler	= proc_dobool,
+	},
 	{
 		.procname	= "ldisc_autoload",
 		.data		= &tty_ldisc_autoload,
Signature · 1 line
2.34.1
Signature · 4 lines
Samuel
---
Pour une évaluation indépendante, transparente et rigoureuse !
Je soutiens la Commission d'Évaluation de l'Inria.
Signature · 4 lines
Samuel
---
Pour une évaluation indépendante, transparente et rigoureuse !
Je soutiens la Commission d'Évaluation de l'Inria.
#6 lore raw
In reply to Samuel Thibault
Quoted 12 lines · Samuel Thibault

Hello,

Kees Cook, le sam. 22 oct. 2022 11:29:49 -0700, a ecrit:

Quoted 8 lines

TIOCSTI continues its long history of being used in privilege escalation attacks[1]. Prior attempts to provide a mechanism to disable this have devolved into discussions around creating full-blown LSMs to provide arbitrary ioctl filtering, which is hugely over-engineered -- only TIOCSTI is being used this way. 3 years ago OpenBSD entirely removed TIOCSTI[2], Android has had it filtered for longer[3], and the tools that had historically used TIOCSTI either do not need it, are not commonly built with it, or have had its use removed.

Quoted 4 lines

No. The Brltty screen reader entirely relies on TIOCSTI to be able to support input from various Braille devices. Please make sure to keep TIOCSTI enabled by default, otherwise some people would just completely lose their usual way of simply typing on Linux.

Yup, it remains default enabled:

Quoted 11 lines

[...]

drivers/tty/Kconfig config LEGACY_PTY_COUNT

Quoted 8 lines
drivers/tty/Kconfig +8 -0
+config LEGACY_TIOCSTI
+	bool "Allow legacy TIOCSTI usage"
+	default y
+	help
+	  Historically the kernel has allowed TIOCSTI, which will push
+	  characters into a controlling TTY. This continues to be used
+	  as a malicious privilege escalation mechanism, and provides no
+	  meaningful real-world utility any more.

Yes it does.

Can you send a patch to adjust this language?

Also, what does FreeBSD use for screen readers?

-Kees

Signature · 1 line
Kees Cook
#7 lore raw

Hello,

Kees Cook, le mar. 27 déc. 2022 19:32:55 -0800, a ecrit:

Quoted 11 lines · Samuel Thibault
Quoted 10 lines

Kees Cook, le sam. 22 oct. 2022 11:29:49 -0700, a ecrit:

Quoted 8 lines

TIOCSTI continues its long history of being used in privilege escalation attacks[1]. Prior attempts to provide a mechanism to disable this have devolved into discussions around creating full-blown LSMs to provide arbitrary ioctl filtering, which is hugely over-engineered -- only TIOCSTI is being used this way. 3 years ago OpenBSD entirely removed TIOCSTI[2], Android has had it filtered for longer[3], and the tools that had historically used TIOCSTI either do not need it, are not commonly built with it, or have had its use removed.

Quoted 6 lines
Quoted 4 lines

No. The Brltty screen reader entirely relies on TIOCSTI to be able to support input from various Braille devices. Please make sure to keep TIOCSTI enabled by default, otherwise some people would just completely lose their usual way of simply typing on Linux.

Yup, it remains default enabled:

Yes, but thining of it, very soon people in various security-sensitive distributions will disable it, as they should indeed. And people who need to use their Braille device on such distributions will get stuck.

Can we perhaps just introduce a CAP_TIOCSTI that the brltty daemon would be able to use? We could even make it only allow TIOCSTI on the linux console (tty->ops == con_ops).

Quoted 1 line

Also, what does FreeBSD use for screen readers?

FreeBSD provides poor support for that, people have to use a patched screen tool to somehow access the console (but only after login).

Samuel

#8 lore raw

Hello,

Samuel Thibault, le mer. 28 déc. 2022 21:57:26 +0100, a ecrit:

Quoted 17 lines

Kees Cook, le mar. 27 déc. 2022 19:32:55 -0800, a ecrit:

Quoted 16 lines · Samuel Thibault
Quoted 14 lines

Kees Cook, le sam. 22 oct. 2022 11:29:49 -0700, a ecrit:

Quoted 8 lines

TIOCSTI continues its long history of being used in privilege escalation attacks[1]. Prior attempts to provide a mechanism to disable this have devolved into discussions around creating full-blown LSMs to provide arbitrary ioctl filtering, which is hugely over-engineered -- only TIOCSTI is being used this way. 3 years ago OpenBSD entirely removed TIOCSTI[2], Android has had it filtered for longer[3], and the tools that had historically used TIOCSTI either do not need it, are not commonly built with it, or have had its use removed.

No. The Brltty screen reader entirely relies on TIOCSTI to be able to support input from various Braille devices. Please make sure to keep TIOCSTI enabled by default, otherwise some people would just completely lose their usual way of simply typing on Linux.

Quoted 5 lines
Quoted 1 line

Yup, it remains default enabled:

Yes, but thining of it, very soon people in various security-sensitive distributions will disable it, as they should indeed. And people who need to use their Braille device on such distributions will get stuck.

And as expected, it did get disabled in Debian for instance, very much to the dismay of blind users, whose keyboard suddenly stopped working at all after rebooting with a Linux 6.3 kernel!...

Quoted 3 lines

Can we perhaps just introduce a CAP_TIOCSTI that the brltty daemon would be able to use? We could even make it only allow TIOCSTI on the linux console (tty->ops == con_ops).

*Please* comment on this so we can progress. ATM people are advising each other to set dev.tty.legacy_tiocsti=1, which is just counter-productive in terms of security...

Really, this a serious regression for the people affected by this.

Samuel

#9 lore raw

Samuel Thibault, le dim. 25 juin 2023 17:56:25 +0200, a ecrit:

Quoted 8 lines

Samuel Thibault, le mer. 28 déc. 2022 21:57:26 +0100, a ecrit:

Quoted 3 lines

Can we perhaps just introduce a CAP_TIOCSTI that the brltty daemon would be able to use? We could even make it only allow TIOCSTI on the linux console (tty->ops == con_ops).

*Please* comment on this so we can progress. ATM people are advising each other to set dev.tty.legacy_tiocsti=1, which is just counter-productive in terms of security...

People are even discussing adding that configuration to the brltty package, which e.g. on ubuntu is installed by default, and thus defeating completely the security measure by default...

Please do contribute to the discussion so we can fix this.

Samuel

#10 lore raw
Quoted 10 lines · Samuel Thibault

Samuel Thibault, le dim. 25 juin 2023 17:56:25 +0200, a ecrit:

Quoted 8 lines

Samuel Thibault, le mer. 28 déc. 2022 21:57:26 +0100, a ecrit:

Quoted 3 lines

Can we perhaps just introduce a CAP_TIOCSTI that the brltty daemon would be able to use? We could even make it only allow TIOCSTI on the linux console (tty->ops == con_ops).

*Please* comment on this so we can progress. ATM people are advising each other to set dev.tty.legacy_tiocsti=1, which is just counter-productive in terms of security...

Quoted 5 lines

People are even discussing adding that configuration to the brltty package, which e.g. on ubuntu is installed by default, and thus defeating completely the security measure by default...

Please do contribute to the discussion so we can fix this.

Hi Samuel,

I'm sorry to hear that this is impacting Braille terminals, but I do believe there are solutions in place which would allow affected users to re-enable TIOCSTI system-wide via the sysctl and then selectively allow access to the terminal applications. However, I do believe they would all require some additional work on the distro/user's part if the user didn't want to continue to allow system-wide access to TIOCSTI.

The first thing that comes to mind is an Android-esque filtering that Kees already mentioned in the commit itself. I'm not an Android expert, but based on the linked "ioctl_macros" file in the Android source, it looks like Android is leveraging the SELinux ioctl extended permission functionality to limit access to TIOCSTI. I'm not sure what experience you have with SELinux, but if you have some understanding of SELinux policy the documentation below might help you get started playing with this:

* https://github.com/SELinuxProject/selinux-notebook/blob/main/src/xperm_rules.md

Another option to restrict TIOCSTI once it has been re-enabled system-wide would be to leverage seccomp to block ioctl(..., TIOCSTI, ...). Sadly, I don't think one would be able to use systemd's existing seccomp filtering as it doesn't support syscall parameters, but I imagine with some work one could add some ioctl smarts to the systemd seccomp code and/or use an existing seccomp sandboxing tool to effectively remove TIOCSTI. Using libseccomp, a simple filter would look something like this (untested pseudocode, you've been warned):

ctx = seccomp_init(SCMP_ACT_ALLOW); seccomp_rule_add(ctx, SCMP_ACT_ERRNO(EIO), SCMP_SYS(ioctl), 1,
SCMP_A1(TIOCSTI));

I'm sure there are some other good ideas that aren't coming to mind right now, but I tend to think that the solutions to this are going to be up in userspace.

Signature · 1 line
paul-moore.com
author
#11 lore raw
In reply to Samuel Thibault
Quoted 3 lines · Samuel Thibault
Quoted 3 lines

Can we perhaps just introduce a CAP_TIOCSTI that the brltty daemon would be able to use? We could even make it only allow TIOCSTI on the linux console (tty->ops == con_ops).

Does brltty run with CAP_SYS_ADMIN? That seems a sensible exception to be made.

Quoted 3 lines

*Please* comment on this so we can progress. ATM people are advising each other to set dev.tty.legacy_tiocsti=1, which is just counter-productive in terms of security...

So is there really no solution for brltty and TIOCSTI being disabled? What is FreeBSD doing? I imagine it's the same situation there too, though maybe there is just no support?

https://www.mail-archive.com/[email protected]/msg02892.html

Quoted 1 line

Really, this a serious regression for the people affected by this.

Can you send a patch adding a CAP_SYS_ADMIN exception?

-Kees

Signature · 1 line
Kees Cook
#12 lore raw

Kees Cook, le mar. 27 juin 2023 19:48:45 -0700, a ecrit:

Quoted 6 lines · Samuel Thibault
Quoted 3 lines
Quoted 3 lines

Can we perhaps just introduce a CAP_TIOCSTI that the brltty daemon would be able to use? We could even make it only allow TIOCSTI on the linux console (tty->ops == con_ops).

Does brltty run with CAP_SYS_ADMIN?

ATM most often, yes, though we are trying to reduce the CAP_* privileges to what it actually needs.

Quoted 5 lines
Quoted 3 lines

*Please* comment on this so we can progress. ATM people are advising each other to set dev.tty.legacy_tiocsti=1, which is just counter-productive in terms of security...

So is there really no solution for brltty and TIOCSTI being disabled?

No, there is no way to simulate characters on the Linux console. The alternative would be to use uinput, but that simulates keycodes, not characters, thus requiring backtranslating first, which is very fragile.

Quoted 2 lines

What is FreeBSD doing? I imagine it's the same situation there too, though maybe there is just no support?

There is just no support in the kernel, only a patch against "screen".

Quoted 3 lines
Quoted 1 line

Really, this a serious regression for the people affected by this.

Can you send a patch adding a CAP_SYS_ADMIN exception?

Sure!

Samuel

author
#13 lore raw
Quoted 24 lines · Samuel Thibault

Kees Cook, le mar. 27 juin 2023 19:48:45 -0700, a ecrit:

Quoted 6 lines · Samuel Thibault
Quoted 3 lines
Quoted 3 lines

Can we perhaps just introduce a CAP_TIOCSTI that the brltty daemon would be able to use? We could even make it only allow TIOCSTI on the linux console (tty->ops == con_ops).

Does brltty run with CAP_SYS_ADMIN?

ATM most often, yes, though we are trying to reduce the CAP_* privileges to what it actually needs.

Quoted 5 lines
Quoted 3 lines

*Please* comment on this so we can progress. ATM people are advising each other to set dev.tty.legacy_tiocsti=1, which is just counter-productive in terms of security...

So is there really no solution for brltty and TIOCSTI being disabled?

No, there is no way to simulate characters on the Linux console. The alternative would be to use uinput, but that simulates keycodes, not characters, thus requiring backtranslating first, which is very fragile.

Quoted 2 lines

What is FreeBSD doing? I imagine it's the same situation there too, though maybe there is just no support?

Quoted 7 lines

There is just no support in the kernel, only a patch against "screen".

Quoted 3 lines
Quoted 1 line

Really, this a serious regression for the people affected by this.

Can you send a patch adding a CAP_SYS_ADMIN exception?

Sure!

Thanks! (And be sure to use file->f_cred for the check[1], not "current", that way brltty can open the tty and drop caps and still do the ioctl.)

-Kees

https://docs.kernel.org/security/credentials.html?highlight=confused+deputy#open-file-credentials

Signature · 1 line
Kees Cook
#14 lore raw

Kees Cook, le mer. 28 juin 2023 09:32:20 -0700, a ecrit:

Quoted 10 lines · Samuel Thibault
Quoted 6 lines

Kees Cook, le mar. 27 juin 2023 19:48:45 -0700, a ecrit:

Quoted 3 lines
Quoted 1 line

Really, this a serious regression for the people affected by this.

Can you send a patch adding a CAP_SYS_ADMIN exception?

Sure!

Thanks! (And be sure to use file->f_cred for the check[1], not "current", that way brltty can open the tty and drop caps and still do the ioctl.)

Actually brltty re-opens the various tty[1-6] consoles when the users switches, so I kept just testing capable(CAP_SYS_ADMIN).

Samuel

author
#15 lore raw
Quoted 7 lines · Samuel Thibault

Kees Cook, le mer. 28 juin 2023 09:32:20 -0700, a ecrit:

Quoted 6 lines · Samuel Thibault
Quoted 5 lines

Kees Cook, le mar. 27 juin 2023 19:48:45 -0700, a ecrit:

Quoted 3 lines
Quoted 1 line

Really, this a serious regression for the people affected by this.

Can you send a patch adding a CAP_SYS_ADMIN exception?

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Sure!

Thanks! (And be sure to use file->f_cred for the check[1], not "current", that way brltty can open the tty and drop caps and still do the ioctl.)

Actually brltty re-opens the various tty[1-6] consoles when the users switches, so I kept just testing capable(CAP_SYS_ADMIN).

Well that's frustrating. :P

Signature · 1 line
Kees Cook
#16 lore raw
In reply to Samuel Thibault

From: Samuel Thibault

Quoted 1 line

Sent: 28 June 2023 07:07

...

Quoted 5 lines
Quoted 1 line

So is there really no solution for brltty and TIOCSTI being disabled?

No, there is no way to simulate characters on the Linux console. The alternative would be to use uinput, but that simulates keycodes, not characters, thus requiring backtranslating first, which is very fragile.

It could probably be rewritten to use a pseudo-tty pair. It might even be possible to emulate (the functionality of) TIOCSTI in the relay process that handles the pseudo-tty.

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

#17 lore raw

David Laight, le jeu. 29 juin 2023 13:23:54 +0000, a ecrit:

Quoted 10 lines

From: Samuel Thibault

Quoted 1 line

Sent: 28 June 2023 07:07

...

Quoted 5 lines
Quoted 1 line

So is there really no solution for brltty and TIOCSTI being disabled?

No, there is no way to simulate characters on the Linux console. The alternative would be to use uinput, but that simulates keycodes, not characters, thus requiring backtranslating first, which is very fragile.

It could probably be rewritten to use a pseudo-tty pair.

That's what yasr does for instance, but that does not make the login prompt accessible, which is a must (just like sighted users don't log in blindly...)

Samuel

author
#18 lore raw
In reply to Kees Cook
Quoted 2 lines · Kees Cook

This is so very long over-due. We just need to do this and put an end to this legacy operation.

thread ping. Should I carry this in my hardening tree instead?

Signature · 1 line
Kees Cook
#19 lore raw
Quoted 5 lines · Kees Cook
Quoted 2 lines · Kees Cook

This is so very long over-due. We just need to do this and put an end to this legacy operation.

thread ping. Should I carry this in my hardening tree instead?

I can take it in my tree, I have a bunch of tty/serial patches to go through this week, give me a few more days please.

thanks,

greg k-h

author
#20 lore raw
Quoted 9 lines · Greg Kroah-Hartman
Quoted 5 lines · Kees Cook
Quoted 2 lines · Kees Cook

This is so very long over-due. We just need to do this and put an end to this legacy operation.

thread ping. Should I carry this in my hardening tree instead?

I can take it in my tree, I have a bunch of tty/serial patches to go through this week, give me a few more days please.

Sounds good; thanks for the update!

Signature · 1 line
Kees Cook