erofs-utils: tar: fix negative GNU base-256 number parsing
PATCH 1/1 erofs-utils: tar: fix negative GNU base-256 number parsing
GNU base-256 fields use a 0xff prefix for negative values, but tarerofs_parsenum() currently accumulates them in signed long long. That does not sign-extend negative values correctly and can also trigger signed-overflow undefined behavior while shifting.
Handle positive and negative GNU base-256 fields separately and do the byte accumulation in unsigned long long instead.
This fixes GNU base-256 decoding for negative tar metadata values such as mtime, uid, gid and device numbers.
--- lib/tar.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-)
lib/tar.c +12 -3
@@ -328,17 +328,26 @@ static long long tarerofs_otoi(const char *ptr, int len)static long long tarerofs_parsenum (const char *ptr,int len) { +const u8 *p = (const u8 *)ptr; + errno =0 ;/* * For fields containing numbers or timestamps that are out of range * for the basic format, the GNU format uses a base-256 representation * instead of an ASCII octal number. */ -if (*(char *)ptr =='\200' || *(char *)ptr =='\377' ) { -long long res =0 ; +if (*(char *)ptr =='\200' ) { +unsigned long long res =0 ;while (--len) - res = (res <<8 ) | (u8)*(++ptr); + res = (res <<8 ) | *(++p); +return res; + } +if (*(char *)ptr =='\377' ) { +unsigned long long res =-1ULL ; + +while (len--) + res = (res <<8 ) | *(p++);return res; }return tarerofs_otoi (ptr, len);
Signature · 1 line
2.43.0
Quoted 5 lines · Vansh Choudhary
GNU base-256 fields use a
0xffprefix for negative values, buttarerofs_parsenum()currently accumulates them in signed long long. That does not sign-extend negative values correctly and can also trigger signed-overflow undefined behavior while shifting.
Quoted 8 lines
Handle positive and negative GNU base-256 fields separately and do the byte accumulation in unsigned long long instead.
This fixes GNU base-256 decoding for negative tar metadata values such as mtime, uid, gid and device numbers.
Fixes 95d315fd7958 ("erofs-utils: introduce tarerofs")Signed-off-by Vansh Choudhary <[email protected]>
Provide a testcase for this?
Thanks,
Gao Xiang
Quoted 38 lines
--- lib/tar.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-)lib/tar.c +12
-3@@ -328,17 +328,26 @@ static long long tarerofs_otoi(const char *ptr, int len)static long long tarerofs_parsenum (const char *ptr,int len) { +const u8 *p = (const u8 *)ptr; + errno =0 ;/* * For fields containing numbers or timestamps that are out of range * for the basic format, the GNU format uses a base-256 representation * instead of an ASCII octal number. */ -if (*(char *)ptr =='\200' || *(char *)ptr =='\377' ) { -long long res =0 ; +if (*(char *)ptr =='\200' ) { +unsigned long long res =0 ;while (--len) - res = (res <<8 ) | (u8)*(++ptr); + res = (res <<8 ) | *(++p); +return res; + } +if (*(char *)ptr =='\377' ) { +unsigned long long res =-1ULL ; + +while (len--) + res = (res <<8 ) | *(p++);return res; }return tarerofs_otoi (ptr, len);
PATCH 1/1 erofs-utils: tests: add test for negative GNU tar mtimes
Add a regression test for negative GNU tar mtimes.
It creates a tarball with a file whose mtime is -1, checks that the mtime field is encoded in GNU base-256 format, and verifies that the timestamp is preserved after building and extracting the image.
--- tests/Makefile.am | 3 +++ tests/erofs/030 | 57 +++++++++++++++++++++++++++++++++++++++++++++ tests/erofs/030.out | 2 ++ 3 files changed, 62 insertions(+) create mode 100755 tests/erofs/030 create mode 100644 tests/erofs/030.out
tests/Makefile.am +3 -0
@@ -126,6 +126,9 @@ TESTS += erofs/028 # 029 - test FUSE daemon and kernel error handling on corrupted inodes TESTS += erofs/029 +# 030 - regression test for negative GNU tar mtimes +TESTS += erofs/030 + # NEW TEST CASE HERE # TESTS += erofs/999
tests/erofs/030 new file +57 -0
57 lines not on this page · raw message
tests/erofs/030.out new file +2 -0
@@ -0,0 +1,2 @@ +QA output created by 030 +Silence is golden
Signature · 1 line
2.43.0
PATCH v2 1/1 erofs-utils: tests: add test for negative GNU tar mtimes
From: Vansh Choudhary <[email protected]>
Add a regression test for negative GNU tar mtimes.
It creates a tarball with a file whose mtime is -1, checks that the mtime field is encoded in GNU base-256 format, and verifies that the timestamp is preserved after building and extracting the image.
---
v2:
- use dump.erofs instead of fsck.erofs;
- switch to MIT license.
tests/Makefile.am | 3 +++ tests/erofs/030 | 55 +++++++++++++++++++++++++++++++++++++++++++++ tests/erofs/030.out | 2 ++ 3 files changed, 60 insertions(+) create mode 100755 tests/erofs/030 create mode 100644 tests/erofs/030.out
tests/Makefile.am +3 -0
@@ -126,6 +126,9 @@ TESTS += erofs/028 # 029 - test FUSE daemon and kernel error handling on corrupted inodes TESTS += erofs/029 +# 030 - regression test for negative GNU tar mtimes +TESTS += erofs/030 + # NEW TEST CASE HERE # TESTS += erofs/999
tests/erofs/030 new file +55 -0
55 lines not on this page · raw message
tests/erofs/030.out new file +2 -0
@@ -0,0 +1,2 @@ +QA output created by 030 +Silence is golden
Signature · 1 line
2.43.5
Quoted 5 lines · Vansh Choudhary
GNU base-256 fields use a
0xffprefix for negative values, buttarerofs_parsenum()currently accumulates them in signed long long. That does not sign-extend negative values correctly and can also trigger signed-overflow undefined behavior while shifting.
Quoted 8 lines
Handle positive and negative GNU base-256 fields separately and do the byte accumulation in unsigned long long instead.
This fixes GNU base-256 decoding for negative tar metadata values such as mtime, uid, gid and device numbers.
Fixes 95d315fd7958 ("erofs-utils: introduce tarerofs")Signed-off-by Vansh Choudhary <[email protected]>
I ran the test but it succeeds, so is it just a UB or it really impacts end users?
Thanks,
Gao Xiang
Quoted 38 lines
--- lib/tar.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-)lib/tar.c +12
-3@@ -328,17 +328,26 @@ static long long tarerofs_otoi(const char *ptr, int len)static long long tarerofs_parsenum (const char *ptr,int len) { +const u8 *p = (const u8 *)ptr; + errno =0 ;/* * For fields containing numbers or timestamps that are out of range * for the basic format, the GNU format uses a base-256 representation * instead of an ASCII octal number. */ -if (*(char *)ptr =='\200' || *(char *)ptr =='\377' ) { -long long res =0 ; +if (*(char *)ptr =='\200' ) { +unsigned long long res =0 ;while (--len) - res = (res <<8 ) | (u8)*(++ptr); + res = (res <<8 ) | *(++p); +return res; + } +if (*(char *)ptr =='\377' ) { +unsigned long long res =-1ULL ; + +while (len--) + res = (res <<8 ) | *(p++);return res; }return tarerofs_otoi (ptr, len);
Yes, this definitely fixes the signed-overflow UB in tarerofs_parsenum(), and also makes the negative GNU base-256 decoding logic correct. But from the current testcase alone, I cannot firmly show a visible end-user failure on the old code path.
So I think this is safest to describe as a correctness + UB fix.
Thanks,
Vansh