Skip to contentSkip to conversation

reiserfs: add check for invalid 1st journal block

PATCH Superseded Pavel Skripkin 2 participants 7 messages May 14, 2021 – Jun 8, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 reiserfs: add check for invalid 1st journal block

syzbot reported divide error in reiserfs.
The problem was in incorrect journal 1st block.

Syzbot's reproducer manualy generated wrong superblock with incorrect 1st block. In journal_init() wasn't any checks about this particular case.

For example, if 1st journal block is before superblock 1st block, it can cause zeroing important superblock members in do_journal_end().

Reported-and-tested-by [email protected]
---
 fs/reiserfs/journal.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)
fs/reiserfs/journal.c +13 -0
index 9edc8e2b154e..d58f24a08385 100644
@@ -2758,6 +2758,19 @@ int journal_init(struct super_block *sb, const char *j_dev_name,
 		goto free_and_return;
 	}
 
+	/*
+	 * Sanity check to see is journal first block correct.
+	 * If journal first block is invalid it can cause
+	 * zeroing important superblock members.
+	 */
+	if (SB_ONDISK_JOURNAL_1st_BLOCK(sb) < SB_JOURNAL_1st_RESERVED_BLOCK(sb)) {
+		reiserfs_warning(sb, "journal-1393",
+			"journal 1st super block is invalid: 1st reserved block %d, but actual 1st block is %d",
+			SB_JOURNAL_1st_RESERVED_BLOCK(sb),
+			SB_ONDISK_JOURNAL_1st_BLOCK(sb));
+		goto free_and_return;
+	}
+
 	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
 		reiserfs_warning(sb, "sh-462",
 				 "unable to initialize journal device");
Signature · 1 line
2.31.1
#1 lore raw
Quoted 7 lines · Pavel Skripkin

syzbot reported divide error in reiserfs.
The problem was in incorrect journal 1st block.

Syzbot's reproducer manualy generated wrong superblock with incorrect 1st block. In journal_init() wasn't any checks about this particular case.

Quoted 6 lines

For example, if 1st journal block is before superblock 1st block, it can cause zeroing important superblock members in do_journal_end().

Reported-and-tested-by [email protected]

Thanks for the patch. One comment below:

Quoted 5 lines
fs/reiserfs/journal.c +0 -0
index 9edc8e2b154e..d58f24a08385 100644

fs/reiserfs/journal.c int journal_init(struct super_block *sb, const char *j_dev_name, · line 2766

Quoted 9 lines
fs/reiserfs/journal.c +6 -0
@@ -2758,6 +2758,19 @@ int journal_init(struct super_block *sb, const char *j_dev_name,
 		goto free_and_return;
 	}
 
+	/*
+	 * Sanity check to see is journal first block correct.
+	 * If journal first block is invalid it can cause
+	 * zeroing important superblock members.
+	 */
+	if (SB_ONDISK_JOURNAL_1st_BLOCK(sb) < SB_JOURNAL_1st_RESERVED_BLOCK(sb)) {

I guess this check is valid only if !SB_ONDISK_JOURNAL_DEVICE(sb), isn't it? Otherwise you are comparing block numbers from two different devices...

Honza

Quoted 13 lines
fs/reiserfs/journal.c +7 -0
+		reiserfs_warning(sb, "journal-1393",
+			"journal 1st super block is invalid: 1st reserved block %d, but actual 1st block is %d",
+			SB_JOURNAL_1st_RESERVED_BLOCK(sb),
+			SB_ONDISK_JOURNAL_1st_BLOCK(sb));
+		goto free_and_return;
+	}
+
 	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
 		reiserfs_warning(sb, "sh-462",
 				 "unable to initialize journal device");
Signature · 1 line
2.31.1
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#2 lore raw
Quoted 26 lines · On Mon, 17 May 2021 12:15:23 +0200 Jan Kara <[email protected]> wrote:
Quoted 7 lines · Pavel Skripkin

syzbot reported divide error in reiserfs.
The problem was in incorrect journal 1st block.

Syzbot's reproducer manualy generated wrong superblock with incorrect 1st block. In journal_init() wasn't any checks about this particular case.

Quoted 7 lines

For example, if 1st journal block is before superblock 1st block, it can cause zeroing important superblock members in do_journal_end().

Reported-and-tested-by:
[email protected]

Thanks for the patch. One comment below:

Quoted 8 lines
fs/reiserfs/journal.c +0 -0
index 9edc8e2b154e..d58f24a08385 100644
@@ -2758,6 +2758,19 @@ int journal_init(struct super_block *sb,
 const char *j_dev_name, goto free_and_return;
 	}
 
Quoted 12 lines

fs/reiserfs/journal.c int journal_init(struct super_block *sb, const char *j_dev_name,

Quoted 7 lines
fs/reiserfs/journal.c +6 -0
+	/*
+	 * Sanity check to see is journal first block correct.
+	 * If journal first block is invalid it can cause
+	 * zeroing important superblock members.
+	 */
+	if (SB_ONDISK_JOURNAL_1st_BLOCK(sb) <
 SB_JOURNAL_1st_RESERVED_BLOCK(sb)) {

I guess this check is valid only if !SB_ONDISK_JOURNAL_DEVICE(sb), isn't it? Otherwise you are comparing block numbers from two different devices...

Hi!

Indeed. Thanks for pointing it out! I'll send v2 soon

Quoted 16 lines

Honza

Quoted 11 lines
fs/reiserfs/journal.c +7 -0
+		reiserfs_warning(sb, "journal-1393",
+			"journal 1st super block is invalid: 1st
 reserved block %d, but actual 1st block is %d",
+			SB_JOURNAL_1st_RESERVED_BLOCK(sb),
+			SB_ONDISK_JOURNAL_1st_BLOCK(sb));
+		goto free_and_return;
+	}
+
 	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
fs/reiserfs/journal.c +0 -0
 		reiserfs_warning(sb, "sh-462",
 				 "unable to initialize journal
Quoted 3 lines

device"); --
2.31.1

With regards,
Pavel Skripkin

#3 lore raw

PATCH v2 1/1 reiserfs: add check for invalid 1st journal block

In reply to Jan Kara

syzbot reported divide error in reiserfs.
The problem was in incorrect journal 1st block.

Syzbot's reproducer manualy generated wrong superblock with incorrect 1st block. In journal_init() wasn't any checks about this particular case.

For example, if 1st journal block is before superblock 1st block, it can cause zeroing important superblock members in do_journal_end().

---

Changes in v2:
Added necessary !SB_ONDISK_JOURNAL_DEVICE(sb) in "if"

---
 fs/reiserfs/journal.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
fs/reiserfs/journal.c +14 -0
index 9edc8e2b154e..0834b101c316 100644
@@ -2758,6 +2758,20 @@ int journal_init(struct super_block *sb, const char *j_dev_name,
 		goto free_and_return;
 	}
 
+	/*
+	 * Sanity check to see if journal first block is correct.
+	 * If journal first block is invalid it can cause
+	 * zeroing important superblock members.
+	 */
+	if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
+	    SB_ONDISK_JOURNAL_1st_BLOCK(sb) < SB_JOURNAL_1st_RESERVED_BLOCK(sb)) {
+		reiserfs_warning(sb, "journal-1393",
+				 "journal 1st super block is invalid: 1st reserved block %d, but actual 1st block is %d",
+				 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
+				 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
+		goto free_and_return;
+	}
+
 	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
 		reiserfs_warning(sb, "sh-462",
 				 "unable to initialize journal device");
Signature · 1 line
2.31.1
#4 lore raw
Quoted 7 lines · Pavel Skripkin

syzbot reported divide error in reiserfs.
The problem was in incorrect journal 1st block.

Syzbot's reproducer manualy generated wrong superblock with incorrect 1st block. In journal_init() wasn't any checks about this particular case.

Quoted 6 lines

For example, if 1st journal block is before superblock 1st block, it can cause zeroing important superblock members in do_journal_end().

Thanks! I've added the patch to my tree.

Honza

Quoted 37 lines

---

Changes in v2:
Added necessary !SB_ONDISK_JOURNAL_DEVICE(sb) in "if"

---
 fs/reiserfs/journal.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
fs/reiserfs/journal.c +14 -0
index 9edc8e2b154e..0834b101c316 100644
@@ -2758,6 +2758,20 @@ int journal_init(struct super_block *sb, const char *j_dev_name,
 		goto free_and_return;
 	}
 
+	/*
+	 * Sanity check to see if journal first block is correct.
+	 * If journal first block is invalid it can cause
+	 * zeroing important superblock members.
+	 */
+	if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
+	    SB_ONDISK_JOURNAL_1st_BLOCK(sb) < SB_JOURNAL_1st_RESERVED_BLOCK(sb)) {
+		reiserfs_warning(sb, "journal-1393",
+				 "journal 1st super block is invalid: 1st reserved block %d, but actual 1st block is %d",
+				 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
+				 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
+		goto free_and_return;
+	}
+
 	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
 		reiserfs_warning(sb, "sh-462",
 				 "unable to initialize journal device");
Signature · 1 line
2.31.1
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#5 lore raw
Quoted 12 lines · On Mon, 17 May 2021 15:08:18 +0200 Jan Kara <[email protected]> wrote:
Quoted 11 lines · Pavel Skripkin

syzbot reported divide error in reiserfs.
The problem was in incorrect journal 1st block.

Syzbot's reproducer manualy generated wrong superblock with incorrect 1st block. In journal_init() wasn't any checks about this particular case.

For example, if 1st journal block is before superblock 1st block, it can cause zeroing important superblock members in do_journal_end().

Quoted 7 lines
Quoted 2 lines

Thanks! I've added the patch to my tree.

Honza

Hi, Jan!

Is this patched got lost somehow? I did't find it in your tree here
https://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs.git/.

Sorry to bother you, maybe Im missing something :)

With regards,
Pavel Skripkin

#6 lore raw
Quoted 23 lines · Pavel Skripkin
Quoted 12 lines · On Mon, 17 May 2021 15:08:18 +0200 Jan Kara <[email protected]> wrote:
Quoted 11 lines · Pavel Skripkin

syzbot reported divide error in reiserfs.
The problem was in incorrect journal 1st block.

Syzbot's reproducer manualy generated wrong superblock with incorrect 1st block. In journal_init() wasn't any checks about this particular case.

For example, if 1st journal block is before superblock 1st block, it can cause zeroing important superblock members in do_journal_end().

Quoted 7 lines
Quoted 2 lines

Thanks! I've added the patch to my tree.

Honza

Quoted 6 lines

Hi, Jan!

Is this patched got lost somehow? I did't find it in your tree here
https://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs.git/.

Sorry to bother you, maybe Im missing something :)

It's sitting in my for_next branch as commit a149127be52f. I'll push it to Linus in the coming merge window.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR