Skip to contentSkip to conversation

libext2fs: Extends commit c9a8c53b, with the same fix for ext2fs_flush2() and ext2fs_image_super_write() on a Big Endian systems.

PATCH In review Anatoly Pugachev 2 participants 3 messages Jan 10, 2020 – Jan 11, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 libext2fs: Extends commit c9a8c53b, with the same fix for ext2fs_flush2() and ext2fs_image_super_write() on a Big Endian systems.

libext2fs: extends commit c9a8c53b, with the same fix for ext2fs_flush2() and ext2fs_image_super_write() on a Big Endian systems.

As follow-up to previous discussion 'dumpe2fs / mke2fs sigserv on sparc64'

Used find for files which refer to:

e2fsprogs.git$ find . -name \*.c | xargs grep -cl 'gdp = ext2fs_group_desc' ./lib/ext2fs/closefs.c
./lib/ext2fs/openfs.c
./lib/ext2fs/imager.c

And applied the same check for a null pointer.

Tested on a debian linux with sparc64 LDOM and ppc64 LPAR.

Fixes sigserv with test suite in "i_bitmaps" test.

---
 lib/ext2fs/closefs.c | 3 ++-
 lib/ext2fs/imager.c  | 6 ++++--
 2 files changed, 6 insertions(+), 3 deletions(-)
lib/ext2fs/closefs.c +2 -1
index 1d4d5b7f..58fdd5c6 100644
@@ -339,7 +339,8 @@ errcode_t ext2fs_flush2(ext2_filsys fs, int flags)
 	ext2fs_swap_super(super_shadow);
 	for (j = 0; j < fs->group_desc_count; j++) {
 		gdp = ext2fs_group_desc(fs, group_shadow, j);
-		ext2fs_swap_group_desc2(fs, gdp);
+		if (gdp)
+			ext2fs_swap_group_desc2(fs, gdp);
 	}
 #else
 	super_shadow = fs->super;
lib/ext2fs/imager.c +4 -2
index 7fd06f74..b40fd826 100644
@@ -245,7 +245,8 @@ errcode_t ext2fs_image_super_write(ext2_filsys fs, int fd,
 	gdp = (struct ext2_group_desc *) cp;
 	for (j=0; j < groups_per_block*fs->desc_blocks; j++) {
 		gdp = ext2fs_group_desc(fs, fs->group_desc, j);
-		ext2fs_swap_group_desc2(fs, gdp);
+		if (gdp)
+			ext2fs_swap_group_desc2(fs, gdp);
 	}
 #endif
 
@@ -257,7 +258,8 @@ errcode_t ext2fs_image_super_write(ext2_filsys fs, int fd,
 	gdp = (struct ext2_group_desc *) cp;
 	for (j=0; j < groups_per_block*fs->desc_blocks; j++) {
 		gdp = ext2fs_group_desc(fs, fs->group_desc, j);
-		ext2fs_swap_group_desc2(fs, gdp);
+		if (gdp)
+			ext2fs_swap_group_desc2(fs, gdp);
 	}
 #endif
 
Signature · 1 line
2.25.0.rc1
#1 lore raw
Quoted 13 lines · Anatoly Pugachev

libext2fs: extends commit c9a8c53b, with the same fix for ext2fs_flush2() and ext2fs_image_super_write() on a Big Endian systems.

As follow-up to previous discussion 'dumpe2fs / mke2fs sigserv on sparc64'

Used find for files which refer to:

e2fsprogs.git$ find . -name \*.c | xargs grep -cl 'gdp = ext2fs_group_desc' ./lib/ext2fs/closefs.c
./lib/ext2fs/openfs.c
./lib/ext2fs/imager.c

Quoted 5 lines

And applied the same check for a null pointer.

Tested on a debian linux with sparc64 LDOM and ppc64 LPAR.

Fixes sigserv with test suite in "i_bitmaps" test.

As far as I know, the i_bitmaps test is passing on on sparc64 and ppc64. Search for i_bitmaps in:

https://buildd.debian.org/status/fetch.php?pkg=e2fsprogs&arch=sparc64&ver=1.45.5-2&stamp=1578527938&raw=0
and
https://buildd.debian.org/status/fetch.php?pkg=e2fsprogs&arch=ppc64&ver=1.45.5-2&stamp=1578526270&raw=0

The bug in c9a8c53b was caused by SPARSE_SUPER being passed to ext2fs_open(). But that doesn't happen in misc/e2image.

I can see optimizing ext2fs_flush() to skip byte-swapping the group descriptors if the SUPER_ONLY flag is enabled. And I can see ext2fs_image_super_write() checking to see if the SUPER_ONLY flag is set, and returning an error in that case.

But I don't think any of the current e2fsprogs are crashing at the moment. Am I missing something?

Regards,

- Ted

#2 lore raw
Quoted 31 lines · Theodore Y. Ts'o
Quoted 13 lines · Anatoly Pugachev

libext2fs: extends commit c9a8c53b, with the same fix for ext2fs_flush2() and ext2fs_image_super_write() on a Big Endian systems.

As follow-up to previous discussion 'dumpe2fs / mke2fs sigserv on sparc64'

Used find for files which refer to:

e2fsprogs.git$ find . -name \*.c | xargs grep -cl 'gdp = ext2fs_group_desc' ./lib/ext2fs/closefs.c
./lib/ext2fs/openfs.c
./lib/ext2fs/imager.c

Quoted 5 lines

And applied the same check for a null pointer.

Tested on a debian linux with sparc64 LDOM and ppc64 LPAR.

Fixes sigserv with test suite in "i_bitmaps" test.

As far as I know, the i_bitmaps test is passing on on sparc64 and ppc64. Search for i_bitmaps in:

https://buildd.debian.org/status/fetch.php?pkg=e2fsprogs&arch=sparc64&ver=1.45.5-2&stamp=1578527938&raw=0
and
https://buildd.debian.org/status/fetch.php?pkg=e2fsprogs&arch=ppc64&ver=1.45.5-2&stamp=1578526270&raw=0

The bug in c9a8c53b was caused by SPARSE_SUPER being passed to ext2fs_open(). But that doesn't happen in misc/e2image.

Quoted 7 lines

I can see optimizing ext2fs_flush() to skip byte-swapping the group descriptors if the SUPER_ONLY flag is enabled. And I can see ext2fs_image_super_write() checking to see if the SUPER_ONLY flag is set, and returning an error in that case.

But I don't think any of the current e2fsprogs are crashing at the moment. Am I missing something?

Ted,

I'm using "master" branch for my tests, and debian probably using "debian/master" .
Even test count is differ. In master branch 366 tests and in "debian/master" 356 tests.

$ git br -vvv
  debian/master 0ba96395 [origin/debian/master] debian/patches: update
for 1.45.5-2 release
* master        32d33132 [origin/master] Merge branch 'maint' into next

Can you please try master branch on any (sparc64 or ppc64) debian porter boxes? (There's new ppc64 porter box coming soon) Or on ppc64/sparc64 "gcc compile test" farm machines?

And it's actually up to you, maybe my patch is irrelevant (and sorry for the noise then), since i don't know internals for e2fsprogs.

Thanks.