Hi,
Using the attached disk image I observe that getdents() never returns
the end of the directory, i.e. mounting the disk image on a loopback
device and running 'ls' under strace shows an endless stream of:
getdents(3, /* 2 entries */, 32768) = 48
getdents(3, /* 2 entries */, 32768) = 48
getdents(3, /* 2 entries */, 32768) = 48
...
Vegard
On Sat, Nov 14, 2015 at 2:19 AM, Vegard Nossum <[email protected]> wrote:
> Hi,
>
> Using the attached disk image I observe that getdents() never returns
> the end of the directory, i.e. mounting the disk image on a loopback
> device and running 'ls' under strace shows an endless stream of:
>
> getdents(3, /* 2 entries */, 32768) = 48
> getdents(3, /* 2 entries */, 32768) = 48
> getdents(3, /* 2 entries */, 32768) = 48
> ...
Please more details. Is this image hand crafted?
If not, how has it been created? Is is supposed to work?
>From a quick look it seems as the root directory is bad but we report
progress in ->iterate.
ctx->pos is 2, we set it back to 0, because of the faked dot entries.
but fat_get_entry() did not make any progress and we report 0 back to VFS.
So, VFS sees progress and the game continues.
Does the attached patch help?
--
Thanks,
//richard
On 11/14/2015 11:32 AM, Richard Weinberger wrote:
> On Sat, Nov 14, 2015 at 2:19 AM, Vegard Nossum <[email protected]> wrote:
>> Hi,
>>
>> Using the attached disk image I observe that getdents() never returns
>> the end of the directory, i.e. mounting the disk image on a loopback
>> device and running 'ls' under strace shows an endless stream of:
>>
>> getdents(3, /* 2 entries */, 32768) = 48
>> getdents(3, /* 2 entries */, 32768) = 48
>> getdents(3, /* 2 entries */, 32768) = 48
>> ...
>
> Please more details. Is this image hand crafted?
> If not, how has it been created? Is is supposed to work?
It was created by fuzzing, it is not supposed to work per se.
> From a quick look it seems as the root directory is bad but we report
> progress in ->iterate.
> ctx->pos is 2, we set it back to 0, because of the faked dot entries.
> but fat_get_entry() did not make any progress and we report 0 back to VFS.
> So, VFS sees progress and the game continues.
>
> Does the attached patch help?
Yes, it does fixes the problem here, but I can't really comment on the
correctness of the patch.
Thanks for the quick reponse,
Vegard
Vegard Nossum <[email protected]> writes:
> On 11/14/2015 11:32 AM, Richard Weinberger wrote:
>> On Sat, Nov 14, 2015 at 2:19 AM, Vegard Nossum <[email protected]> wrote:
>>> Hi,
>>>
>>> Using the attached disk image I observe that getdents() never returns
>>> the end of the directory, i.e. mounting the disk image on a loopback
>>> device and running 'ls' under strace shows an endless stream of:
>>>
>>> getdents(3, /* 2 entries */, 32768) = 48
>>> getdents(3, /* 2 entries */, 32768) = 48
>>> getdents(3, /* 2 entries */, 32768) = 48
>>> ...
>>
>> Please more details. Is this image hand crafted?
>> If not, how has it been created? Is is supposed to work?
>
> It was created by fuzzing, it is not supposed to work per se.
>
>> From a quick look it seems as the root directory is bad but we report
>> progress in ->iterate.
>> ctx->pos is 2, we set it back to 0, because of the faked dot entries.
>> but fat_get_entry() did not make any progress and we report 0 back to VFS.
>> So, VFS sees progress and the game continues.
>>
>> Does the attached patch help?
>
> Yes, it does fixes the problem here, but I can't really comment on the
> correctness of the patch.
>
> Thanks for the quick reponse,
I made cleanup and made sure fake_offset is corrected.
Richard, Signed-off-by was missed in your patch, so I added. Can you
agree to Signed-off-by?
Thanks.
--
OGAWA Hirofumi <[email protected]>
[PATCH] fat: Fix fake_offset handling on error path
[[email protected]: cleanup and make sure to correct fake_offset]
Signed-off-by: Richard Weinberger <[email protected]>
Signed-off-by: OGAWA Hirofumi <[email protected]>
---
fs/fat/dir.c | 25 +++++++++++++++----------
1 file changed, 15 insertions(+), 10 deletions(-)
diff -puN fs/fat/dir.c~fat-fake_offset-fix fs/fat/dir.c
--- linux-tux3/fs/fat/dir.c~fat-fake_offset-fix 2015-11-14 21:53:21.722269967 +0900
+++ linux-tux3-hirofumi/fs/fat/dir.c 2015-11-14 23:03:16.740608521 +0900
@@ -559,7 +559,7 @@ static int __fat_readdir(struct inode *i
{
struct super_block *sb = inode->i_sb;
struct msdos_sb_info *sbi = MSDOS_SB(sb);
- struct buffer_head *bh;
+ struct buffer_head *bh = NULL;
struct msdos_dir_entry *de;
unsigned char nr_slots;
wchar_t *unicode = NULL;
@@ -588,10 +588,9 @@ static int __fat_readdir(struct inode *i
goto out;
}
- bh = NULL;
get_new:
if (fat_get_entry(inode, &cpos, &bh, &de) == -1)
- goto end_of_dir;
+ goto out;
parse_record:
nr_slots = 0;
/*
@@ -614,7 +613,7 @@ parse_record:
int status = fat_parse_long(inode, &cpos, &bh, &de,
&unicode, &nr_slots);
if (status < 0) {
- ctx->pos = cpos;
+ bh = NULL;
ret = status;
goto out;
} else if (status == PARSE_INVALID)
@@ -622,7 +621,7 @@ parse_record:
else if (status == PARSE_NOT_LONGNAME)
goto parse_record;
else if (status == PARSE_EOF)
- goto end_of_dir;
+ goto out;
if (nr_slots) {
void *longname = unicode + FAT_MAX_UNI_CHARS;
@@ -658,8 +657,9 @@ parse_record:
fill_len = short_len;
start_filldir:
- if (!fake_offset)
- ctx->pos = cpos - (nr_slots + 1) * sizeof(struct msdos_dir_entry);
+ ctx->pos = cpos - (nr_slots + 1) * sizeof(struct msdos_dir_entry);
+ if (fake_offset && ctx->pos < 2)
+ ctx->pos = 2;
if (!memcmp(de->name, MSDOS_DOT, MSDOS_NAME)) {
if (!dir_emit_dot(file, ctx))
@@ -685,14 +685,19 @@ record_end:
fake_offset = 0;
ctx->pos = cpos;
goto get_new;
-end_of_dir:
- ctx->pos = cpos;
+
+out:
+ if (fake_offset && cpos < 2)
+ ctx->pos = 2;
+ else
+ ctx->pos = cpos;
fill_failed:
brelse(bh);
if (unicode)
__putname(unicode);
-out:
+
mutex_unlock(&sbi->s_lock);
+
return ret;
}
_
Am 14.11.2015 um 15:28 schrieb OGAWA Hirofumi:
> Vegard Nossum <[email protected]> writes:
>
>> On 11/14/2015 11:32 AM, Richard Weinberger wrote:
>>> On Sat, Nov 14, 2015 at 2:19 AM, Vegard Nossum <[email protected]> wrote:
>>>> Hi,
>>>>
>>>> Using the attached disk image I observe that getdents() never returns
>>>> the end of the directory, i.e. mounting the disk image on a loopback
>>>> device and running 'ls' under strace shows an endless stream of:
>>>>
>>>> getdents(3, /* 2 entries */, 32768) = 48
>>>> getdents(3, /* 2 entries */, 32768) = 48
>>>> getdents(3, /* 2 entries */, 32768) = 48
>>>> ...
>>>
>>> Please more details. Is this image hand crafted?
>>> If not, how has it been created? Is is supposed to work?
>>
>> It was created by fuzzing, it is not supposed to work per se.
>>
>>> From a quick look it seems as the root directory is bad but we report
>>> progress in ->iterate.
>>> ctx->pos is 2, we set it back to 0, because of the faked dot entries.
>>> but fat_get_entry() did not make any progress and we report 0 back to VFS.
>>> So, VFS sees progress and the game continues.
>>>
>>> Does the attached patch help?
>>
>> Yes, it does fixes the problem here, but I can't really comment on the
>> correctness of the patch.
>>
>> Thanks for the quick reponse,
>
> I made cleanup and made sure fake_offset is corrected.
>
> Richard, Signed-off-by was missed in your patch, so I added. Can you
> agree to Signed-off-by?
Sure!
Thanks,
//richard
Richard Weinberger <[email protected]> writes:
>>> Yes, it does fixes the problem here, but I can't really comment on the
>>> correctness of the patch.
>>>
>>> Thanks for the quick reponse,
>>
>> I made cleanup and made sure fake_offset is corrected.
>>
>> Richard, Signed-off-by was missed in your patch, so I added. Can you
>> agree to Signed-off-by?
>
> Sure!
Attached updated patch made smaller changes, and with missed Cc: stable.
Andrew, please queue this up.
Thanks.
--
OGAWA Hirofumi <[email protected]>
[PATCH] fat: Fix fake_offset handling on error path
[[email protected]: cleanup and make sure to correct fake_offset]
Cc: [email protected]
Signed-off-by: Richard Weinberger <[email protected]>
Signed-off-by: OGAWA Hirofumi <[email protected]>
---
fs/fat/dir.c | 16 +++++++++++-----
1 file changed, 11 insertions(+), 5 deletions(-)
diff -puN fs/fat/dir.c~fat-fake_offset-fix fs/fat/dir.c
--- linux/fs/fat/dir.c~fat-fake_offset-fix 2015-11-14 23:31:45.904700155 +0900
+++ linux-hirofumi/fs/fat/dir.c 2015-11-15 02:45:13.861256766 +0900
@@ -610,9 +610,9 @@ parse_record:
int status = fat_parse_long(inode, &cpos, &bh, &de,
&unicode, &nr_slots);
if (status < 0) {
- ctx->pos = cpos;
+ bh = NULL;
ret = status;
- goto out;
+ goto end_of_dir;
} else if (status == PARSE_INVALID)
goto record_end;
else if (status == PARSE_NOT_LONGNAME)
@@ -654,8 +654,9 @@ parse_record:
fill_len = short_len;
start_filldir:
- if (!fake_offset)
- ctx->pos = cpos - (nr_slots + 1) * sizeof(struct msdos_dir_entry);
+ ctx->pos = cpos - (nr_slots + 1) * sizeof(struct msdos_dir_entry);
+ if (fake_offset && ctx->pos < 2)
+ ctx->pos = 2;
if (!memcmp(de->name, MSDOS_DOT, MSDOS_NAME)) {
if (!dir_emit_dot(file, ctx))
@@ -681,14 +682,19 @@ record_end:
fake_offset = 0;
ctx->pos = cpos;
goto get_new;
+
end_of_dir:
- ctx->pos = cpos;
+ if (fake_offset && cpos < 2)
+ ctx->pos = 2;
+ else
+ ctx->pos = cpos;
fill_failed:
brelse(bh);
if (unicode)
__putname(unicode);
out:
mutex_unlock(&sbi->s_lock);
+
return ret;
}
_
On 11/14/2015 07:19 PM, OGAWA Hirofumi wrote:
> Richard Weinberger <[email protected]> writes:
>
>>>> Yes, it does fixes the problem here, but I can't really comment on the
>>>> correctness of the patch.
>>>>
>>>> Thanks for the quick reponse,
>>>
>>> I made cleanup and made sure fake_offset is corrected.
>>>
>>> Richard, Signed-off-by was missed in your patch, so I added. Can you
>>> agree to Signed-off-by?
>>
>> Sure!
>
> Attached updated patch made smaller changes, and with missed Cc: stable.
>
> Andrew, please queue this up.
>
> Thanks.
>
It would be nice to have a proper patch description too. How about this?
"""
For the root directory, . and .. are faked (using dir_emit_dots()) and
ctx->pos is reset from 2 to 0.
A corrupted root directory could cause fat_get_entry() to fail, but
->iterate() (fat_readdir()) reports progress to the VFS (with ctx->pos
rewound to 0), so any following calls to ->iterate() continue to return
the same entries again and again.
The result is that userspace will never see the end of the directory,
causing e.g. 'ls' to hang in a getdents() loop.
"""
Thanks,
Vegard
Am 15.11.2015 um 12:05 schrieb Vegard Nossum:
> On 11/14/2015 07:19 PM, OGAWA Hirofumi wrote:
>> Richard Weinberger <[email protected]> writes:
>>
>>>>> Yes, it does fixes the problem here, but I can't really comment on the
>>>>> correctness of the patch.
>>>>>
>>>>> Thanks for the quick reponse,
>>>>
>>>> I made cleanup and made sure fake_offset is corrected.
>>>>
>>>> Richard, Signed-off-by was missed in your patch, so I added. Can you
>>>> agree to Signed-off-by?
>>>
>>> Sure!
>>
>> Attached updated patch made smaller changes, and with missed Cc: stable.
>>
>> Andrew, please queue this up.
>>
>> Thanks.
>>
>
> It would be nice to have a proper patch description too. How about this?
>
> """
> For the root directory, . and .. are faked (using dir_emit_dots()) and
> ctx->pos is reset from 2 to 0.
>
> A corrupted root directory could cause fat_get_entry() to fail, but
> ->iterate() (fat_readdir()) reports progress to the VFS (with ctx->pos
> rewound to 0), so any following calls to ->iterate() continue to return
> the same entries again and again.
>
> The result is that userspace will never see the end of the directory,
> causing e.g. 'ls' to hang in a getdents() loop.
> """
Agreed. And you deserve also a Reported-and-tested-by. :-)
Thanks,
//richard
Richard Weinberger <[email protected]> writes:
>> It would be nice to have a proper patch description too. How about this?
>>
>> """
>> For the root directory, . and .. are faked (using dir_emit_dots()) and
>> ctx->pos is reset from 2 to 0.
>>
>> A corrupted root directory could cause fat_get_entry() to fail, but
>> ->iterate() (fat_readdir()) reports progress to the VFS (with ctx->pos
>> rewound to 0), so any following calls to ->iterate() continue to return
>> the same entries again and again.
>>
>> The result is that userspace will never see the end of the directory,
>> causing e.g. 'ls' to hang in a getdents() loop.
>> """
>
> Agreed. And you deserve also a Reported-and-tested-by. :-)
Sounds good, updated patch here.
--
OGAWA Hirofumi <[email protected]>
[PATCH v3] fat: Fix fake_offset handling on error path
For the root directory, . and .. are faked (using dir_emit_dots()) and
ctx->pos is reset from 2 to 0.
A corrupted root directory could cause fat_get_entry() to fail, but
->iterate() (fat_readdir()) reports progress to the VFS (with ctx->pos
rewound to 0), so any following calls to ->iterate() continue to return
the same entries again and again.
The result is that userspace will never see the end of the directory,
causing e.g. 'ls' to hang in a getdents() loop.
[[email protected]: cleanup and make sure to correct fake_offset]
Cc: [email protected]
Reported-and-tested-by: Vegard Nossum <[email protected]>
Signed-off-by: Richard Weinberger <[email protected]>
Signed-off-by: OGAWA Hirofumi <[email protected]>
---
fs/fat/dir.c | 16 +++++++++++-----
1 file changed, 11 insertions(+), 5 deletions(-)
diff -puN fs/fat/dir.c~fat-fake_offset-fix fs/fat/dir.c
--- linux/fs/fat/dir.c~fat-fake_offset-fix 2015-11-14 23:31:45.904700155 +0900
+++ linux-hirofumi/fs/fat/dir.c 2015-11-15 02:45:13.861256766 +0900
@@ -610,9 +610,9 @@ parse_record:
int status = fat_parse_long(inode, &cpos, &bh, &de,
&unicode, &nr_slots);
if (status < 0) {
- ctx->pos = cpos;
+ bh = NULL;
ret = status;
- goto out;
+ goto end_of_dir;
} else if (status == PARSE_INVALID)
goto record_end;
else if (status == PARSE_NOT_LONGNAME)
@@ -654,8 +654,9 @@ parse_record:
fill_len = short_len;
start_filldir:
- if (!fake_offset)
- ctx->pos = cpos - (nr_slots + 1) * sizeof(struct msdos_dir_entry);
+ ctx->pos = cpos - (nr_slots + 1) * sizeof(struct msdos_dir_entry);
+ if (fake_offset && ctx->pos < 2)
+ ctx->pos = 2;
if (!memcmp(de->name, MSDOS_DOT, MSDOS_NAME)) {
if (!dir_emit_dot(file, ctx))
@@ -681,14 +682,19 @@ record_end:
fake_offset = 0;
ctx->pos = cpos;
goto get_new;
+
end_of_dir:
- ctx->pos = cpos;
+ if (fake_offset && cpos < 2)
+ ctx->pos = 2;
+ else
+ ctx->pos = cpos;
fill_failed:
brelse(bh);
if (unicode)
__putname(unicode);
out:
mutex_unlock(&sbi->s_lock);
+
return ret;
}
_