2018-02-26 15:49:28

by Ilya Smith

[permalink] [raw]
Subject: [PATCH 0/1] Additional strict check on ELF file.

Hello, this patch implements strict check on segments order in ELF file.
Current implementation doesn't check the order of ELF segments and compute
totalsize as difference between last one and the first one. If ELF file
has wrong order it allows to re-mmap existing memory region with one
from ELF file.

Ilya Smith (1):
Additional strict check on ELF file. Checks segments are followed in
order of 'p_vaddr ' value ascending. It fixes erorr in
total_mapping_size with computation total size. This error happens
if segments in ELF file are not in order.

fs/binfmt_elf.c | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 51 insertions(+)

--
2.14.1



2018-02-26 15:49:53

by Ilya Smith

[permalink] [raw]
Subject: [PATCH 1/1] Additional strict check on ELF file. Checks segments are followed in order of 'p_vaddr ' value ascending. It fixes erorr in total_mapping_size with computation total size. This error happens if segments in ELF file are not in order.

Signed-off-by: Ilya Smith <[email protected]>
---
fs/binfmt_elf.c | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 51 insertions(+)

diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index bdb201230bae..970b42044240 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c
@@ -524,6 +524,52 @@ static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,

#endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */

+/**
+ * elf_check_phdr() - common check ELF program header.
+ * @phdr: The program header to check
+ * @phdr_num: Count of program headers in @phdr from elf header.
+ *
+ * Checks ELF binary meets specification.
+ *
+ * Return: Zero to proceed with ELF load, non-zero to faile the ELF load
+ * with that return code.
+ */
+static int elf_check_phdr(struct elf_phdr *phdr, unsigned long phdr_num)
+{
+ unsigned long i;
+ struct elf_phdr *eppnt = phdr;
+ Elf64_Addr curr_vaddr;
+ Elf64_Xword curr_memsz;
+
+ /* Find first PT_LOAD entry */
+ for (i = 0; i < phdr_num && eppnt->p_type != PT_LOAD; ++i, ++eppnt)
+ ;
+
+ /* no any PT_LOAD */
+ if (i == phdr_num)
+ return -EINVAL;
+
+ curr_memsz = eppnt->p_memsz;
+ curr_vaddr = eppnt->p_vaddr;
+
+ for (++i, ++eppnt; i < phdr_num; ++i, ++eppnt) {
+ if (eppnt->p_type != PT_LOAD)
+ continue;
+
+ /* Check order of vaddr */
+ if (eppnt->p_vaddr <= curr_vaddr)
+ return -EINVAL;
+
+ /* Check overlapping */
+ if (eppnt->p_vaddr < curr_vaddr + curr_memsz)
+ return -EINVAL;
+
+ curr_memsz = eppnt->p_memsz;
+ curr_vaddr = eppnt->p_vaddr;
+ }
+ return 0;
+}
+
/* This is much more generalized than the library routine read function,
so we keep this separate. Technically the library read function
is only provided so that we can read a.out libraries that have
@@ -551,6 +597,8 @@ static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
goto out;
if (!interpreter->f_op->mmap)
goto out;
+ if (elf_check_phdr(interp_elf_phdata, interp_elf_ex->e_phnum))
+ goto out;

total_size = total_mapping_size(interp_elf_phdata,
interp_elf_ex->e_phnum);
@@ -733,6 +781,9 @@ static int load_elf_binary(struct linux_binprm *bprm)
if (!elf_phdata)
goto out;

+ if (elf_check_phdr(&loc->elf_ex, loc->elf_ex.e_phnum))
+ goto out;
+
elf_ppnt = elf_phdata;
elf_bss = 0;
elf_brk = 0;
--
2.14.1


2018-02-26 17:50:33

by Randy Dunlap

[permalink] [raw]
Subject: Re: [PATCH 1/1] Additional strict check on ELF file. Checks segments are followed in order of 'p_vaddr ' value ascending. It fixes erorr in total_mapping_size with computation total size. This error happens if segments in ELF file are not in order.

On 02/26/2018 07:46 AM, Ilya Smith wrote:
> Signed-off-by: Ilya Smith <[email protected]>
> ---
> fs/binfmt_elf.c | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 51 insertions(+)
>
> diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> index bdb201230bae..970b42044240 100644
> --- a/fs/binfmt_elf.c
> +++ b/fs/binfmt_elf.c
> @@ -524,6 +524,52 @@ static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
>
> #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
>
> +/**
> + * elf_check_phdr() - common check ELF program header.
> + * @phdr: The program header to check
> + * @phdr_num: Count of program headers in @phdr from elf header.
> + *
> + * Checks ELF binary meets specification.
> + *
> + * Return: Zero to proceed with ELF load, non-zero to faile the ELF load

fail

> + * with that return code.
> + */
> +static int elf_check_phdr(struct elf_phdr *phdr, unsigned long phdr_num)
> +{

And it would be nicer/better to have all of the "intro" text in the body of this
message instead of in another email.

thanks,
--
~Randy

2018-02-27 12:54:18

by Ilya Smith

[permalink] [raw]
Subject: Re: [PATCH 1/1] Additional strict check on ELF file. Checks segments are followed in order of 'p_vaddr ' value ascending. It fixes erorr in total_mapping_size with computation total size. This error happens if segments in ELF file are not in order.

Thanks for that, I will fix it in next patch. Any other feedback?

Thanks,
Ilya

> 26 февр. 2018 г., в 20:48, Randy Dunlap <[email protected]> написал(а):
>
> On 02/26/2018 07:46 AM, Ilya Smith wrote:
>> Signed-off-by: Ilya Smith <[email protected]>
>> ---
>> fs/binfmt_elf.c | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++
>> 1 file changed, 51 insertions(+)
>>
>> diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
>> index bdb201230bae..970b42044240 100644
>> --- a/fs/binfmt_elf.c
>> +++ b/fs/binfmt_elf.c
>> @@ -524,6 +524,52 @@ static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
>>
>> #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
>>
>> +/**
>> + * elf_check_phdr() - common check ELF program header.
>> + * @phdr: The program header to check
>> + * @phdr_num: Count of program headers in @phdr from elf header.
>> + *
>> + * Checks ELF binary meets specification.
>> + *
>> + * Return: Zero to proceed with ELF load, non-zero to faile the ELF load
>
> fail
>
>> + * with that return code.
>> + */
>> +static int elf_check_phdr(struct elf_phdr *phdr, unsigned long phdr_num)
>> +{
>
> And it would be nicer/better to have all of the "intro" text in the body of this
> message instead of in another email.
>
> thanks,
> --
> ~Randy