Skip to contentSkip to conversation

ubifs: avoid assert failed in ubifs.c

PATCH New Patrice Chotard 5 participants 6 messages Apr 27, 2018 – May 22, 2018 UBI file system (UBIFS)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ubifs: avoid assert failed in ubifs.c

This patch solves assert failed displayed in the console during a boot. The root cause is that the ubifs_inode is not already allocated when ubifs_printdir and ubifs_finddir functions are called.

Trace showing the issue:
feed 'boot.scr.uimg', ino 94, new f_pos 0x17b40ece dent->ch.sqnum '7132', creat_sqnum 3886945402880 UBIFS assert failed in ubifs_finddir at 436 INODE ALLOCATION: creat_sqnum '7129'
Found U-Boot script /boot.scr.uimg

---

 fs/ubifs/ubifs.c | 4 ++++
 1 file changed, 4 insertions(+)
fs/ubifs/ubifs.c +4 -0
index 4465523d5fbe..a71c9400ef33 100644
@@ -350,7 +350,9 @@ static int ubifs_printdir(struct file *file, void *dirent)
 		dbg_gen("feed '%s', ino %llu, new f_pos %#x",
 			dent->name, (unsigned long long)le64_to_cpu(dent->inum),
 			key_hash_flash(c, &dent->key));
+#ifndef __UBOOT__
 		ubifs_assert(le64_to_cpu(dent->ch.sqnum) > ubifs_inode(dir)->creat_sqnum);
+#endif
 
 		nm.len = le16_to_cpu(dent->nlen);
 		over = filldir(c, (char *)dent->name, nm.len,
@@ -432,7 +434,9 @@ static int ubifs_finddir(struct super_block *sb, char *dirname,
 		dbg_gen("feed '%s', ino %llu, new f_pos %#x",
 			dent->name, (unsigned long long)le64_to_cpu(dent->inum),
 			key_hash_flash(c, &dent->key));
+#ifndef __UBOOT__
 		ubifs_assert(le64_to_cpu(dent->ch.sqnum) > ubifs_inode(dir)->creat_sqnum);
+#endif
 
 		nm.len = le16_to_cpu(dent->nlen);
 		if ((strncmp(dirname, (char *)dent->name, nm.len) == 0) &&
Signature · 1 line
1.9.1
#1 lore raw

PATCH 1/1 ubifs: avoid assert failed in ubifs.c

Hello Patrice,

Am 27.04.2018 um 15:51 schrieb Patrice Chotard:

Quoted 11 lines

This patch solves assert failed displayed in the console during a boot. The root cause is that the ubifs_inode is not already allocated when ubifs_printdir and ubifs_finddir functions are called.

Trace showing the issue:
feed 'boot.scr.uimg', ino 94, new f_pos 0x17b40ece dent->ch.sqnum '7132', creat_sqnum 3886945402880 UBIFS assert failed in ubifs_finddir at 436 INODE ALLOCATION: creat_sqnum '7129'
Found U-Boot script /boot.scr.uimg

Quoted 6 lines

---

  fs/ubifs/ubifs.c | 4 ++++
  1 file changed, 4 insertions(+)

Applied to u-boot-ubi.git master

Thanks!

bye,
Heiko

Signature · 3 lines
DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-52   Fax: +49-8142-66989-80   Email: hs at denx.de
#2 lore raw

PATCH 1/1 ubifs: avoid assert failed in ubifs.c

In reply to Patrice Chotard
Quoted 11 lines · Patrice Chotard

This patch solves assert failed displayed in the console during a boot. The root cause is that the ubifs_inode is not already allocated when ubifs_printdir and ubifs_finddir functions are called.

Trace showing the issue:
feed 'boot.scr.uimg', ino 94, new f_pos 0x17b40ece dent->ch.sqnum '7132', creat_sqnum 3886945402880 UBIFS assert failed in ubifs_finddir at 436 INODE ALLOCATION: creat_sqnum '7129'
Found U-Boot script /boot.scr.uimg

Quoted 2 lines

I ran into this too, but what I do not quite understand from the commit message is how hiding the error actually solves the problem that the assert points to.

Why does the assert trigger in the first place ?

What is the root cause of the issue that is being hidden by this patch?

Quoted 30 lines

---

 fs/ubifs/ubifs.c | 4 ++++
 1 file changed, 4 insertions(+)
fs/ubifs/ubifs.c +4 -0
index 4465523d5fbe..a71c9400ef33 100644
@@ -350,7 +350,9 @@ static int ubifs_printdir(struct file *file, void *dirent)
 		dbg_gen("feed '%s', ino %llu, new f_pos %#x",
 			dent->name, (unsigned long long)le64_to_cpu(dent->inum),
 			key_hash_flash(c, &dent->key));
+#ifndef __UBOOT__
 		ubifs_assert(le64_to_cpu(dent->ch.sqnum) > ubifs_inode(dir)->creat_sqnum);
+#endif
 
 		nm.len = le16_to_cpu(dent->nlen);
 		over = filldir(c, (char *)dent->name, nm.len,
@@ -432,7 +434,9 @@ static int ubifs_finddir(struct super_block *sb, char *dirname,
 		dbg_gen("feed '%s', ino %llu, new f_pos %#x",
 			dent->name, (unsigned long long)le64_to_cpu(dent->inum),
 			key_hash_flash(c, &dent->key));
+#ifndef __UBOOT__
 		ubifs_assert(le64_to_cpu(dent->ch.sqnum) > ubifs_inode(dir)->creat_sqnum);
+#endif
 
 		nm.len = le16_to_cpu(dent->nlen);
 		if ((strncmp(dirname, (char *)dent->name, nm.len) == 0) &&
Signature · 2 lines
Best regards,
Marek Vasut
#3 lore raw

PATCH 1/1 ubifs: avoid assert failed in ubifs.c

Quoted 15 lines · Marek Vasut
Quoted 11 lines · Patrice Chotard

This patch solves assert failed displayed in the console during a boot. The root cause is that the ubifs_inode is not already allocated when ubifs_printdir and ubifs_finddir functions are called.

Trace showing the issue:
feed 'boot.scr.uimg', ino 94, new f_pos 0x17b40ece dent->ch.sqnum '7132', creat_sqnum 3886945402880 UBIFS assert failed in ubifs_finddir at 436 INODE ALLOCATION: creat_sqnum '7129'
Found U-Boot script /boot.scr.uimg

Quoted 2 lines
Quoted 7 lines

I ran into this too, but what I do not quite understand from the commit message is how hiding the error actually solves the problem that the assert points to.

Why does the assert trigger in the first place ?

What is the root cause of the issue that is being hidden by this patch?

Bump?

Quoted 32 lines
Quoted 30 lines

---

 fs/ubifs/ubifs.c | 4 ++++
 1 file changed, 4 insertions(+)
fs/ubifs/ubifs.c +4 -0
index 4465523d5fbe..a71c9400ef33 100644
@@ -350,7 +350,9 @@ static int ubifs_printdir(struct file *file, void *dirent)
 		dbg_gen("feed '%s', ino %llu, new f_pos %#x",
 			dent->name, (unsigned long long)le64_to_cpu(dent->inum),
 			key_hash_flash(c, &dent->key));
+#ifndef __UBOOT__
 		ubifs_assert(le64_to_cpu(dent->ch.sqnum) > ubifs_inode(dir)->creat_sqnum);
+#endif
 
 		nm.len = le16_to_cpu(dent->nlen);
 		over = filldir(c, (char *)dent->name, nm.len,
@@ -432,7 +434,9 @@ static int ubifs_finddir(struct super_block *sb, char *dirname,
 		dbg_gen("feed '%s', ino %llu, new f_pos %#x",
 			dent->name, (unsigned long long)le64_to_cpu(dent->inum),
 			key_hash_flash(c, &dent->key));
+#ifndef __UBOOT__
 		ubifs_assert(le64_to_cpu(dent->ch.sqnum) > ubifs_inode(dir)->creat_sqnum);
+#endif
 
 		nm.len = le16_to_cpu(dent->nlen);
 		if ((strncmp(dirname, (char *)dent->name, nm.len) == 0) &&
Signature · 2 lines
Best regards,
Marek Vasut
#4 lore raw

PATCH 1/1 ubifs: avoid assert failed in ubifs.c

Am Dienstag, 22. Mai 2018, 12:56:48 CEST schrieb Marek Vasut:

Quoted 20 lines · Marek Vasut
Quoted 19 lines · Patrice Chotard
Quoted 11 lines

This patch solves assert failed displayed in the console during a boot. The root cause is that the ubifs_inode is not already allocated when ubifs_printdir and ubifs_finddir functions are called.

Trace showing the issue:
feed 'boot.scr.uimg', ino 94, new f_pos 0x17b40ece dent->ch.sqnum '7132', creat_sqnum 3886945402880 UBIFS assert failed in ubifs_finddir at 436 INODE ALLOCATION: creat_sqnum '7129'
Found U-Boot script /boot.scr.uimg

Quoted 2 lines

I ran into this too, but what I do not quite understand from the commit message is how hiding the error actually solves the problem that the assert points to.

Quoted 5 lines
Quoted 3 lines

Why does the assert trigger in the first place ?

What is the root cause of the issue that is being hidden by this patch?

Bump?

I had a look, the bug is deeper, ubifs_finddir() allocates a vfs inode manually and ignores UBIFS internals. ubifs_inode() will read beyond the allocated buffer. In best case the assert triggers because ->creat_sqnum is garbage, in worst case, U-Boot will just crash.

AFAICT, the correct solution is to use ubifs_iget(). Then we can keep the assert and it will check for the right thing.

Thanks,
//richard

#5 lore raw

PATCH 1/1 ubifs: avoid assert failed in ubifs.c

Quoted 21 lines · Richard Weinberger

Am Dienstag, 22. Mai 2018, 12:56:48 CEST schrieb Marek Vasut:

Quoted 20 lines · Marek Vasut
Quoted 19 lines · Patrice Chotard
Quoted 11 lines

This patch solves assert failed displayed in the console during a boot. The root cause is that the ubifs_inode is not already allocated when ubifs_printdir and ubifs_finddir functions are called.

Trace showing the issue:
feed 'boot.scr.uimg', ino 94, new f_pos 0x17b40ece dent->ch.sqnum '7132', creat_sqnum 3886945402880 UBIFS assert failed in ubifs_finddir at 436 INODE ALLOCATION: creat_sqnum '7129'
Found U-Boot script /boot.scr.uimg

Quoted 2 lines

I ran into this too, but what I do not quite understand from the commit message is how hiding the error actually solves the problem that the assert points to.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Why does the assert trigger in the first place ?

What is the root cause of the issue that is being hidden by this patch?

Bump?

I had a look, the bug is deeper, ubifs_finddir() allocates a vfs inode manually

                                   ^^^^^^^^^^^^^^^

Hi Richard,

above triggered alert, so here's my old attempt to simplify ubifs_finddir(). Things could be further simplified, but I got distracted and never had time to look here again (which I'm affraid of).

Subject: [PATCH] Refactor ubifs_finddir

Just work in progress...

---
 fs/ubifs/ubifs.c | 67 +++++++-----------------------------------------
 1 file changed, 9 insertions(+), 58 deletions(-)
fs/ubifs/ubifs.c +9 -58
index 47fa41ad1d..a876546a0e 100644
@@ -389,47 +389,20 @@ out:
 static int ubifs_finddir(struct super_block *sb, char *dirname,
 			 unsigned long root_inum, unsigned long *inum)
 {
-	int err;
 	struct qstr nm;
 	union ubifs_key key;
 	struct ubifs_dent_node *dent;
-	struct ubifs_info *c;
-	struct file *file;
-	struct dentry *dentry;
-	struct inode *dir;
-	int ret = 0;
-
-	file = kzalloc(sizeof(struct file), 0);
-	dentry = kzalloc(sizeof(struct dentry), 0);
-	dir = kzalloc(sizeof(struct inode), 0);
-	if (!file || !dentry || !dir) {
-		printf("%s: Error, no memory for malloc!\n", __func__);
-		err = -ENOMEM;
-		goto out;
-	}
-
-	dir->i_sb = sb;
-	file->f_path.dentry = dentry;
-	file->f_path.dentry->d_parent = dentry;
-	file->f_path.dentry->d_inode = dir;
-	file->f_path.dentry->d_inode->i_ino = root_inum;
-	c = sb->s_fs_info;
-
-	dbg_gen("dir ino %lu, f_pos %#llx", dir->i_ino, file->f_pos);
+	struct ubifs_info *c = sb->s_fs_info;
 
 	/* Find the first entry in TNC and save it */
-	lowest_dent_key(c, &key, dir->i_ino);
+	lowest_dent_key(c, &key, root_inum);
 	nm.name = NULL;
-	dent = ubifs_tnc_next_ent(c, &key, &nm);
-	if (IS_ERR(dent)) {
-		err = PTR_ERR(dent);
-		goto out;
-	}
-
-	file->f_pos = key_hash_flash(c, &dent->key);
-	file->private_data = dent;
 
 	while (1) {
+		dent = ubifs_tnc_next_ent(c, &key, &nm);
+		if (IS_ERR(dent))
+			return PTR_ERR(dent);
+
 		dbg_gen("feed '%s', ino %llu, new f_pos %#x",
 			dent->name, (unsigned long long)le64_to_cpu(dent->inum),
 			key_hash_flash(c, &dent->key));
@@ -441,36 +414,14 @@ static int ubifs_finddir(struct super_block *sb, char *dirname,
 		if ((strncmp(dirname, (char *)dent->name, nm.len) == 0) &&
 		    (strlen(dirname) == nm.len)) {
 			*inum = le64_to_cpu(dent->inum);
-			ret = 1;
-			goto out_free;
+			kfree(dent);
+			return 0;
 		}
-
 		/* Switch to the next entry */
 		key_read(c, &dent->key, &key);
 		nm.name = (char *)dent->name;
-		dent = ubifs_tnc_next_ent(c, &key, &nm);
-		if (IS_ERR(dent)) {
-			err = PTR_ERR(dent);
-			goto out;
-		}
-
-		kfree(file->private_data);
-		file->f_pos = key_hash_flash(c, &dent->key);
-		file->private_data = dent;
-		cond_resched();
+		kfree(dent);
 	}
-
-out:
-	if (err != -ENOENT)
-		dbg_gen("cannot find next direntry, error %d", err);
-
-out_free:
-	kfree(file->private_data);
-	free(file);
-	free(dentry);
-	free(dir);
-
-	return ret;
 }
 
 static unsigned long ubifs_findfile(struct super_block *sb, char *filename)
Signature · 1 line
2.17.0
Quoted 13 lines

and ignores UBIFS internals. ubifs_inode() will read beyond the allocated buffer. In best case the assert triggers because ->creat_sqnum is garbage, in worst case, U-Boot will just crash.

AFAICT, the correct solution is to use ubifs_iget(). Then we can keep the assert and it will check for the right thing.

Thanks,
//richard
_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot