Skip to contentSkip to conversation

procfs: block chmod on /proc/thread-self/comm

PATCH In review Aleksa Sarai 4 participants 5 messages Jul 13, 2023

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 procfs: block chmod on /proc/thread-self/comm

Due to an oversight in commit 1b3044e39a89 ("procfs: fix pthread cross-thread naming if !PR_DUMPABLE") in switching from REG to NOD, chmod operations on /proc/thread-self/comm were no longer blocked as they are on almost all other procfs files.

A very similar situation with /proc/self/environ was used to as a root exploit a long time ago, but procfs has SB_I_NOEXEC so this is simply a correctness issue.

Ref: https://lwn.net/Articles/191954/
Ref: 6d76fa58b050 ("Don't allow chmod() on the /proc/<pid>/ files")

Fixes 1b3044e39a89 ("procfs: fix pthread cross-thread naming if !PR_DUMPABLE")
---
 fs/proc/base.c                               | 3 ++-
 tools/testing/selftests/nolibc/nolibc-test.c | 4 ++++
 2 files changed, 6 insertions(+), 1 deletion(-)
fs/proc/base.c +2 -1
index 05452c3b9872..7394229816f3 100644
@@ -3583,7 +3583,8 @@ static int proc_tid_comm_permission(struct mnt_idmap *idmap,
 }
 
 static const struct inode_operations proc_tid_comm_inode_operations = {
-		.permission = proc_tid_comm_permission,
+		.setattr	= proc_setattr,
+		.permission	= proc_tid_comm_permission,
 };
 
 /*
tools/testing/selftests/nolibc/nolibc-test.c +4 -0
index 486334981e60..08f0969208eb 100644
@@ -580,6 +580,10 @@ int run_syscall(int min, int max)
 		CASE_TEST(chmod_net);         EXPECT_SYSZR(proc, chmod("/proc/self/net", 0555)); break;
 		CASE_TEST(chmod_self);        EXPECT_SYSER(proc, chmod("/proc/self", 0555), -1, EPERM); break;
 		CASE_TEST(chown_self);        EXPECT_SYSER(proc, chown("/proc/self", 0, 0), -1, EPERM); break;
+		CASE_TEST(chmod_self_comm);   EXPECT_SYSER(proc, chmod("/proc/self/comm", 0777), -1, EPERM); break;
+		CASE_TEST(chmod_tid_comm);    EXPECT_SYSER(proc, chmod("/proc/thread-self/comm", 0777), -1, EPERM); break;
+		CASE_TEST(chmod_self_environ);EXPECT_SYSER(proc, chmod("/proc/self/environ", 0777), -1, EPERM); break;
+		CASE_TEST(chmod_tid_environ); EXPECT_SYSER(proc, chmod("/proc/thread-self/environ", 0777), -1, EPERM); break;
 		CASE_TEST(chroot_root);       EXPECT_SYSZR(euid0, chroot("/")); break;
 		CASE_TEST(chroot_blah);       EXPECT_SYSER(1, chroot("/proc/self/blah"), -1, ENOENT); break;
 		CASE_TEST(chroot_exe);        EXPECT_SYSER(proc, chroot("/proc/self/exe"), -1, ENOTDIR); break;
Signature · 1 line
2.41.0
maintainer
#1 lore raw

+Cc Thomas Weißschuh <[email protected]> as this seems quite related to his finding about /proc/self/net:

https://lore.kernel.org/lkml/[email protected]/#b

Quoted 50 lines · Aleksa Sarai

Due to an oversight in commit 1b3044e39a89 ("procfs: fix pthread cross-thread naming if !PR_DUMPABLE") in switching from REG to NOD, chmod operations on /proc/thread-self/comm were no longer blocked as they are on almost all other procfs files.

A very similar situation with /proc/self/environ was used to as a root exploit a long time ago, but procfs has SB_I_NOEXEC so this is simply a correctness issue.

Ref: https://lwn.net/Articles/191954/
Ref: 6d76fa58b050 ("Don't allow chmod() on the /proc/<pid>/ files")

Fixes 1b3044e39a89 ("procfs: fix pthread cross-thread naming if !PR_DUMPABLE")
---
 fs/proc/base.c                               | 3 ++-
 tools/testing/selftests/nolibc/nolibc-test.c | 4 ++++
 2 files changed, 6 insertions(+), 1 deletion(-)
fs/proc/base.c +2 -1
index 05452c3b9872..7394229816f3 100644
@@ -3583,7 +3583,8 @@ static int proc_tid_comm_permission(struct mnt_idmap *idmap,
 }
 
 static const struct inode_operations proc_tid_comm_inode_operations = {
-		.permission = proc_tid_comm_permission,
+		.setattr	= proc_setattr,
+		.permission	= proc_tid_comm_permission,
 };
 
 /*
tools/testing/selftests/nolibc/nolibc-test.c +4 -0
index 486334981e60..08f0969208eb 100644
@@ -580,6 +580,10 @@ int run_syscall(int min, int max)
 		CASE_TEST(chmod_net);         EXPECT_SYSZR(proc, chmod("/proc/self/net", 0555)); break;
 		CASE_TEST(chmod_self);        EXPECT_SYSER(proc, chmod("/proc/self", 0555), -1, EPERM); break;
 		CASE_TEST(chown_self);        EXPECT_SYSER(proc, chown("/proc/self", 0, 0), -1, EPERM); break;
+		CASE_TEST(chmod_self_comm);   EXPECT_SYSER(proc, chmod("/proc/self/comm", 0777), -1, EPERM); break;
+		CASE_TEST(chmod_tid_comm);    EXPECT_SYSER(proc, chmod("/proc/thread-self/comm", 0777), -1, EPERM); break;
+		CASE_TEST(chmod_self_environ);EXPECT_SYSER(proc, chmod("/proc/self/environ", 0777), -1, EPERM); break;
+		CASE_TEST(chmod_tid_environ); EXPECT_SYSER(proc, chmod("/proc/thread-self/environ", 0777), -1, EPERM); break;
 		CASE_TEST(chroot_root);       EXPECT_SYSZR(euid0, chroot("/")); break;
 		CASE_TEST(chroot_blah);       EXPECT_SYSER(1, chroot("/proc/self/blah"), -1, ENOENT); break;
 		CASE_TEST(chroot_exe);        EXPECT_SYSER(proc, chroot("/proc/self/exe"), -1, ENOTDIR); break;
Signature · 1 line
2.41.0
#2 lore raw
Quoted 4 lines · Willy Tarreau

+Cc Thomas Weißschuh <[email protected]> as this seems quite related to his finding about /proc/self/net:

https://lore.kernel.org/lkml/[email protected]/#b

Yeah I saw this patch and (along with an earlier discussion with Christian on the topic of chmod on symlinks -- see [1]) lead us to find that there were three other cases where this happens unintentionally:

* /proc/self (on the symlink itself)
* /proc/thread-self (on the symlink itself)
* /proc/thread-self/comm

The first two will be fixed by [1] so fixing them isn't necessary.

[1]: https://lore.kernel.org/linux-fsdevel/[email protected]/

Signature · 4 lines
Aleksa Sarai
Senior Software Engineer (Containers)
SUSE Linux GmbH
<https://www.cyphar.com/>
#3 lore raw
In reply to Aleksa Sarai
Quoted 24 lines · Aleksa Sarai

Due to an oversight in commit 1b3044e39a89 ("procfs: fix pthread cross-thread naming if !PR_DUMPABLE") in switching from REG to NOD, chmod operations on /proc/thread-self/comm were no longer blocked as they are on almost all other procfs files.

A very similar situation with /proc/self/environ was used to as a root exploit a long time ago, but procfs has SB_I_NOEXEC so this is simply a correctness issue.

Ref: https://lwn.net/Articles/191954/
Ref: 6d76fa58b050 ("Don't allow chmod() on the /proc/<pid>/ files")

Fixes 1b3044e39a89 ("procfs: fix pthread cross-thread naming if !PR_DUMPABLE")
---
 fs/proc/base.c                               | 3 ++-
 tools/testing/selftests/nolibc/nolibc-test.c | 4 ++++
 2 files changed, 6 insertions(+), 1 deletion(-)
fs/proc/base.c +0 -0
index 05452c3b9872..7394229816f3 100644

fs/proc/base.c static int proc_tid_comm_permission(struct mnt_idmap *idmap, · line 3588

Quoted 7 lines
fs/proc/base.c +2 -1
@@ -3583,7 +3583,8 @@ static int proc_tid_comm_permission(struct mnt_idmap *idmap,
 }
 
 static const struct inode_operations proc_tid_comm_inode_operations = {
-		.permission = proc_tid_comm_permission,
+		.setattr	= proc_setattr,
+		.permission	= proc_tid_comm_permission,
 };

Given that this seems to be a recurring theme a more systematic aproach would help.

Something like the following (untested) patch:

fs/proc/base.c +1 -0
index 05452c3b9872..b90f2e9cda66 100644
@@ -2649,6 +2649,7 @@ static struct dentry *proc_pident_instantiate(struct dentry *dentry,
 		set_nlink(inode, 2);	/* Use getattr to fix if necessary */
 	if (p->iop)
 		inode->i_op = p->iop;
+	WARN_ON(!inode->i_op->setattr);
 	if (p->fop)
 		inode->i_fop = p->fop;
 	ei->op = p->op;

tools/testing/selftests/nolibc/nolibc-test.c int run_syscall(int min, int max) · line 586

Quoted 13 lines
fs/proc/base.c +0 -0
 /*
tools/testing/selftests/nolibc/nolibc-test.c +4 -0
index 486334981e60..08f0969208eb 100644
@@ -580,6 +580,10 @@ int run_syscall(int min, int max)
 		CASE_TEST(chmod_net);         EXPECT_SYSZR(proc, chmod("/proc/self/net", 0555)); break;
 		CASE_TEST(chmod_self);        EXPECT_SYSER(proc, chmod("/proc/self", 0555), -1, EPERM); break;
 		CASE_TEST(chown_self);        EXPECT_SYSER(proc, chown("/proc/self", 0, 0), -1, EPERM); break;
+		CASE_TEST(chmod_self_comm);   EXPECT_SYSER(proc, chmod("/proc/self/comm", 0777), -1, EPERM); break;
+		CASE_TEST(chmod_tid_comm);    EXPECT_SYSER(proc, chmod("/proc/thread-self/comm", 0777), -1, EPERM); break;
+		CASE_TEST(chmod_self_environ);EXPECT_SYSER(proc, chmod("/proc/self/environ", 0777), -1, EPERM); break;
+		CASE_TEST(chmod_tid_environ); EXPECT_SYSER(proc, chmod("/proc/thread-self/environ", 0777), -1, EPERM); break;

I'm not a big fan of this, it abuses the nolibc testsuite to test core kernel functionality.
If this needs to be tested explicitly there is hopefully a better place.

Those existing tests focus on testing functionality provided by nolibc. The test chmod_net just got removed because it suffered from the same bug as /proc/thread-self/comm.

Quoted 6 lines
tools/testing/selftests/nolibc/nolibc-test.c +0 -0
 		CASE_TEST(chroot_root);       EXPECT_SYSZR(euid0, chroot("/")); break;
 		CASE_TEST(chroot_blah);       EXPECT_SYSER(1, chroot("/proc/self/blah"), -1, ENOENT); break;
 		CASE_TEST(chroot_exe);        EXPECT_SYSER(proc, chroot("/proc/self/exe"), -1, ENOTDIR); break;
Signature · 1 line
2.41.0
#4 lore raw
Quoted 43 lines · Thomas Weißschuh
Quoted 24 lines · Aleksa Sarai

Due to an oversight in commit 1b3044e39a89 ("procfs: fix pthread cross-thread naming if !PR_DUMPABLE") in switching from REG to NOD, chmod operations on /proc/thread-self/comm were no longer blocked as they are on almost all other procfs files.

A very similar situation with /proc/self/environ was used to as a root exploit a long time ago, but procfs has SB_I_NOEXEC so this is simply a correctness issue.

Ref: https://lwn.net/Articles/191954/
Ref: 6d76fa58b050 ("Don't allow chmod() on the /proc/<pid>/ files")

Fixes 1b3044e39a89 ("procfs: fix pthread cross-thread naming if !PR_DUMPABLE")
---
 fs/proc/base.c                               | 3 ++-
 tools/testing/selftests/nolibc/nolibc-test.c | 4 ++++
 2 files changed, 6 insertions(+), 1 deletion(-)
fs/proc/base.c +0 -0
index 05452c3b9872..7394229816f3 100644

fs/proc/base.c static int proc_tid_comm_permission(struct mnt_idmap *idmap, · line 3588

Quoted 7 lines
fs/proc/base.c +2 -1
@@ -3583,7 +3583,8 @@ static int proc_tid_comm_permission(struct mnt_idmap *idmap,
 }
 
 static const struct inode_operations proc_tid_comm_inode_operations = {
-		.permission = proc_tid_comm_permission,
+		.setattr	= proc_setattr,
+		.permission	= proc_tid_comm_permission,
 };

Given that this seems to be a recurring theme a more systematic aproach would help.

Something like the following (untested) patch:

fs/proc/base.c +0 -0
index 05452c3b9872..b90f2e9cda66 100644

fs/proc/base.c static struct dentry *proc_pident_instantiate(struct dentry *dentry, · line 2652

Quoted 4 lines
fs/proc/base.c +1 -0
@@ -2649,6 +2649,7 @@ static struct dentry *proc_pident_instantiate(struct dentry *dentry,
 		set_nlink(inode, 2);	/* Use getattr to fix if necessary */
 	if (p->iop)
 		inode->i_op = p->iop;
+	WARN_ON(!inode->i_op->setattr);

Hm, no. This is hacky.

To fix this properly we will need to wean off notify_change() from falling back to simple_setattr() when no i_op->setattr() method is defined. To do that we will have to go through every filesystem and port all that rely on this fallback to set simple_setattr() explicitly as their i_op->setattr() method.

Christoph and I just discussed this in relation to another patch.

This is a bugfix so it should be as minimal as possible for easy backport.