Skip to contentSkip to conversation

erofs-utils: fix undefined behavior shift in erofs_init_devices

PATCH In review Nithurshen 2 participants 6 messages Apr 19 – Apr 20

Subsystems (2)
lore
Share & tools
lore

Nithurshen

author
#0 lore raw

PATCH 1/2 erofs-utils: fix undefined behavior shift in erofs_init_devices

In erofs_init_devices(), roundup_pow_of_two() can potentially trigger an undefined behavior shift if the incremented 'ondisk_extradevs' value results in an overflow or an input that leads to an out-of-bounds shift.

Promote the argument to u64 before the increment to ensure the rounding logic operates on a safe bit-width.

---
 lib/super.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/super.c +1 -1
index 088c9a0..10831a7 100644
@@ -49,7 +49,7 @@ static int erofs_init_devices(struct erofs_sb_info *sbi,
 		return 0;
 
 	sbi->extra_devices = ondisk_extradevs;
-	sbi->device_id_mask = roundup_pow_of_two(ondisk_extradevs + 1) - 1;
+	sbi->device_id_mask = roundup_pow_of_two((u64)ondisk_extradevs + 1) - 1;
 	sbi->devs = calloc(ondisk_extradevs, sizeof(*sbi->devs));
 	if (!sbi->devs)
 		return -ENOMEM;
Signature · 1 line
2.52.0
author
#1 lore raw

PATCH 2/2 erofs-utils: libzstd: fix undefined behavior shift in setdictsize

In erofs_compressor_libzstd_setdictsize(), if pclustersize_max is 0, dict_size becomes 0, leading to undefined behavior when calling ilog2(0). This results in an invalid bit shift (e.g., shifting a 32-bit value by 63 bits), as reported by cppcheck.

Fix this by adding guards to ensure dict_size is non-zero before performing power-of-two rounding and validation.

---
 lib/compressor_libzstd.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
lib/compressor_libzstd.c +3 -2
index 6330f44..eb768de 100644
@@ -123,10 +123,11 @@ static int erofs_compressor_libzstd_setdictsize(struct erofs_compress *c,
 		} else {
 			dict_size = min_t(u32, Z_EROFS_ZSTD_MAX_DICT_SIZE,
 					  pclustersize_max << 3);
-			dict_size = 1U << ilog2(dict_size);
+			if (dict_size)
+				dict_size = 1U << ilog2(dict_size);
 		}
 	}
-	if (dict_size != 1U << ilog2(dict_size) ||
+	if (!dict_size || dict_size != 1U << ilog2(dict_size) ||
 	    dict_size > Z_EROFS_ZSTD_MAX_DICT_SIZE) {
 		erofs_err("invalid dictionary size %u", dict_size);
 		return -EINVAL;
Signature · 1 line
2.52.0
#2 lore raw

Re: [PATCH 2/2] erofs-utils: libzstd: fix undefined behavior shift in setdictsize

Quoted 1 line · Nithurshen

In erofs_compressor_libzstd_setdictsize(), if pclustersize_max is 0,

How pclustersize_max is 0? explain please?

#3 lore raw
In reply to Nithurshen
Quoted 4 lines · Nithurshen

In erofs_init_devices(), roundup_pow_of_two() can potentially trigger an undefined behavior shift if the incremented 'ondisk_extradevs' value results in an overflow or an input that leads to an out-of-bounds shift.

I wonder why there is "out-of-bounds shift"? why you all think it's an issue? Can you please explain in details?

ondisk_extradevs is 65535 at most.

Thanks,
Gao Xiang

Quoted 22 lines

Promote the argument to u64 before the increment to ensure the rounding logic operates on a safe bit-width.

---
  lib/super.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
lib/super.c +1 -1
index 088c9a0..10831a7 100644
@@ -49,7 +49,7 @@ static int erofs_init_devices(struct erofs_sb_info *sbi,
  		return 0;
 
  	sbi->extra_devices = ondisk_extradevs;
-	sbi->device_id_mask = roundup_pow_of_two(ondisk_extradevs + 1) - 1;
+	sbi->device_id_mask = roundup_pow_of_two((u64)ondisk_extradevs + 1) - 1;
  	sbi->devs = calloc(ondisk_extradevs, sizeof(*sbi->devs));
  	if (!sbi->devs)
  		return -ENOMEM;
author
#4 lore raw

PATCH 2/2 erofs-utils: libzstd: fix undefined behavior shift in setdictsize

Hi Xiang,

I do agree that pclustersize_max is always greater than 0 and ondisk_extradevs is at max 65535.

My intent was to look at these functions as independent blocks. Even though the current upstream callers provide safe inputs, I wanted to make them robust against any undefined behavior in case they are reused in the future.

Thanks,
Nithurshen

Regards,
Nithurshen

#5 lore raw

Re: [PATCH 2/2] erofs-utils: libzstd: fix undefined behavior shift in setdictsize

Quoted 9 lines · Nithurshen

Hi Xiang,

I do agree that pclustersize_max is always greater than 0 and ondisk_extradevs is at max 65535.

My intent was to look at these functions as independent blocks. Even though the current upstream callers provide safe inputs, I wanted to make them robust against any undefined behavior in case they are reused in the future.

They won't be reused in any case like this.

Quoted 6 lines

Thanks,
Nithurshen

Regards,
Nithurshen