Re: [PATCH v1] kexec/arch/i386: Add support for KASLR memory randomization

[Simon Horman]

On Tue, Sep 13, 2016 at 03:10:05PM +0800, Baoquan He wrote:
> From: Thomas Garnier 
> 
> Multiple changes were made on KASLR (right now in linux-next). One of
> them is randomizing the virtual address of the physical mapping, vmalloc
> and vmemmap memory sections. It breaks kdump ability to read physical
> memory.
> 
> This change identifies if KASLR memories randomization is used by
> checking if the page_offset_base variable exists. It search for the
> correct PAGE_OFFSET value by looking at the loaded memory section and
> find the lowest aligned on PUD (the randomization level).
> 
> Related commits on linux-next:
>  - 0483e1fa6e09d4948272680f691dccb1edb9677f: Base for randomization
>  - 021182e52fe01c1f7b126f97fd6ba048dc4234fd: Enable for PAGE_OFFSET
> 
> Signed-off-by: Thomas Garnier 

Thanks, applied.

___
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

Re: [PATCH v1] kexec/arch/i386: Add support for KASLR memory randomization

[Thomas Garnier]

On Thu, Sep 22, 2016 at 1:41 AM, Dave Young  wrote:
> Hi, Thomas
>
> On 08/17/16 at 09:47am, Thomas Garnier wrote:
>> Multiple changes were made on KASLR (right now in linux-next). One of
>> them is randomizing the virtual address of the physical mapping, vmalloc
>> and vmemmap memory sections. It breaks kdump ability to read physical
>> memory.
>
> What is the user visible behavior without this patch? Could you add more
> in the patch log?
>
> During my testing seems with or without this patch kdump kernel boot
> both fine.

Without this patch, you can't access memory on the generated crash dumps.

>
> My kernel config options is like below, is it enough to test this patch?
> CONFIG_RANDOMIZE_BASE=y
> CONFIG_X86_NEED_RELOCS=y
> CONFIG_PHYSICAL_ALIGN=0x100
> CONFIG_RANDOMIZE_MEMORY=y
> CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING=0x0
>

Should be good enough.

>>
>> This change identifies if KASLR memories randomization is used by
>> checking if the page_offset_base variable exists. It search for the
>> correct PAGE_OFFSET value by looking at the loaded memory section and
>> find the lowest aligned on PUD (the randomization level).
>>
>> Related commits on linux-next:
>>  - 0483e1fa6e09d4948272680f691dccb1edb9677f: Base for randomization
>>  - 021182e52fe01c1f7b126f97fd6ba048dc4234fd: Enable for PAGE_OFFSET
>>
>> Signed-off-by: Thomas Garnier 
>> ---
>>  kexec/arch/i386/crashdump-x86.c | 29 ++---
>>  1 file changed, 22 insertions(+), 7 deletions(-)
>>
>> diff --git a/kexec/arch/i386/crashdump-x86.c 
>> b/kexec/arch/i386/crashdump-x86.c
>> index bbc0f35..ab833d4 100644
>> --- a/kexec/arch/i386/crashdump-x86.c
>> +++ b/kexec/arch/i386/crashdump-x86.c
>> @@ -102,11 +102,10 @@ static int get_kernel_paddr(struct kexec_info 
>> *UNUSED(info),
>>   return -1;
>>  }
>>
>> -/* Retrieve kernel _stext symbol virtual address from /proc/kallsyms */
>> -static unsigned long long get_kernel_stext_sym(void)
>> +/* Retrieve kernel symbol virtual address from /proc/kallsyms */
>> +static unsigned long long get_kernel_sym(const char *symbol)
>
> It sounds better to split this to another patch.
>

Why not, that's a very small change though.

>>  {
>>   const char *kallsyms = "/proc/kallsyms";
>> - const char *stext = "_stext";
>>   char sym[128];
>>   char line[128];
>>   FILE *fp;
>> @@ -122,13 +121,13 @@ static unsigned long long get_kernel_stext_sym(void)
>>   while(fgets(line, sizeof(line), fp) != NULL) {
>>   if (sscanf(line, "%Lx %c %s", , , sym) != 3)
>>   continue;
>> - if (strcmp(sym, stext) == 0) {
>> - dbgprintf("kernel symbol %s vaddr = %16llx\n", stext, 
>> vaddr);
>> + if (strcmp(sym, symbol) == 0) {
>> + dbgprintf("kernel symbol %s vaddr = %16llx\n", symbol, 
>> vaddr);
>>   return vaddr;
>>   }
>>   }
>>
>> - fprintf(stderr, "Cannot get kernel %s symbol address\n", stext);
>> + fprintf(stderr, "Cannot get kernel %s symbol address\n", symbol);
>>   return 0;
>>  }
>>
>> @@ -151,6 +150,8 @@ static int get_kernel_vaddr_and_size(struct kexec_info 
>> *UNUSED(info),
>>   off_t size;
>>   uint32_t elf_flags = 0;
>>   uint64_t stext_sym;
>> + const unsigned long long pud_mask = ~((1 << 30) - 1);
>> + unsigned long long vaddr, lowest_vaddr = 0;
>>
>>   if (elf_info->machine != EM_X86_64)
>>   return 0;
>> @@ -180,9 +181,23 @@ static int get_kernel_vaddr_and_size(struct kexec_info 
>> *UNUSED(info),
>>
>>   end_phdr = _phdr[ehdr.e_phnum];
>>
>> + /* Search for the real PAGE_OFFSET when KASLR memory randomization
>> +  * is enabled */
>> + if (get_kernel_sym("page_offset_base") != 0) {
>> + for(phdr = ehdr.e_phdr; phdr != end_phdr; phdr++) {
>> + if (phdr->p_type == PT_LOAD) {
>> + vaddr = phdr->p_vaddr & pud_mask;
>> + if (lowest_vaddr == 0 || lowest_vaddr > vaddr)
>> + lowest_vaddr = vaddr;
>> + }
>> + }
>> + if (lowest_vaddr != 0)
>> + elf_info->page_offset = lowest_vaddr;
>> + }
>> +
>>   /* Traverse through the Elf headers and find the region where
>>* _stext symbol is located in. That's where kernel is mapped */
>> - stext_sym = get_kernel_stext_sym();
>> + stext_sym = get_kernel_sym("_stext");
>>   for(phdr = ehdr.e_phdr; stext_sym && phdr != end_phdr; phdr++) {
>>   if (phdr->p_type == PT_LOAD) {
>>   unsigned long long saddr = phdr->p_vaddr;
>> --
>> 2.8.0.rc3.226.g39d4020
>>
>>
>> ___
>> kexec mailing list
>> kexec@lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/kexec
>
> Thanks
> Dave

Re: [PATCH v1] kexec/arch/i386: Add support for KASLR memory randomization

[Dave Young]

Hi, Thomas

On 08/17/16 at 09:47am, Thomas Garnier wrote:
> Multiple changes were made on KASLR (right now in linux-next). One of
> them is randomizing the virtual address of the physical mapping, vmalloc
> and vmemmap memory sections. It breaks kdump ability to read physical
> memory.

What is the user visible behavior without this patch? Could you add more
in the patch log?

During my testing seems with or without this patch kdump kernel boot
both fine.

My kernel config options is like below, is it enough to test this patch?
CONFIG_RANDOMIZE_BASE=y
CONFIG_X86_NEED_RELOCS=y
CONFIG_PHYSICAL_ALIGN=0x100
CONFIG_RANDOMIZE_MEMORY=y
CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING=0x0

> 
> This change identifies if KASLR memories randomization is used by
> checking if the page_offset_base variable exists. It search for the
> correct PAGE_OFFSET value by looking at the loaded memory section and
> find the lowest aligned on PUD (the randomization level).
> 
> Related commits on linux-next:
>  - 0483e1fa6e09d4948272680f691dccb1edb9677f: Base for randomization
>  - 021182e52fe01c1f7b126f97fd6ba048dc4234fd: Enable for PAGE_OFFSET
> 
> Signed-off-by: Thomas Garnier 
> ---
>  kexec/arch/i386/crashdump-x86.c | 29 ++---
>  1 file changed, 22 insertions(+), 7 deletions(-)
> 
> diff --git a/kexec/arch/i386/crashdump-x86.c b/kexec/arch/i386/crashdump-x86.c
> index bbc0f35..ab833d4 100644
> --- a/kexec/arch/i386/crashdump-x86.c
> +++ b/kexec/arch/i386/crashdump-x86.c
> @@ -102,11 +102,10 @@ static int get_kernel_paddr(struct kexec_info 
> *UNUSED(info),
>   return -1;
>  }
>  
> -/* Retrieve kernel _stext symbol virtual address from /proc/kallsyms */
> -static unsigned long long get_kernel_stext_sym(void)
> +/* Retrieve kernel symbol virtual address from /proc/kallsyms */
> +static unsigned long long get_kernel_sym(const char *symbol)

It sounds better to split this to another patch.

>  {
>   const char *kallsyms = "/proc/kallsyms";
> - const char *stext = "_stext";
>   char sym[128];
>   char line[128];
>   FILE *fp;
> @@ -122,13 +121,13 @@ static unsigned long long get_kernel_stext_sym(void)
>   while(fgets(line, sizeof(line), fp) != NULL) {
>   if (sscanf(line, "%Lx %c %s", , , sym) != 3)
>   continue;
> - if (strcmp(sym, stext) == 0) {
> - dbgprintf("kernel symbol %s vaddr = %16llx\n", stext, 
> vaddr);
> + if (strcmp(sym, symbol) == 0) {
> + dbgprintf("kernel symbol %s vaddr = %16llx\n", symbol, 
> vaddr);
>   return vaddr;
>   }
>   }
>  
> - fprintf(stderr, "Cannot get kernel %s symbol address\n", stext);
> + fprintf(stderr, "Cannot get kernel %s symbol address\n", symbol);
>   return 0;
>  }
>  
> @@ -151,6 +150,8 @@ static int get_kernel_vaddr_and_size(struct kexec_info 
> *UNUSED(info),
>   off_t size;
>   uint32_t elf_flags = 0;
>   uint64_t stext_sym;
> + const unsigned long long pud_mask = ~((1 << 30) - 1);
> + unsigned long long vaddr, lowest_vaddr = 0;
>  
>   if (elf_info->machine != EM_X86_64)
>   return 0;
> @@ -180,9 +181,23 @@ static int get_kernel_vaddr_and_size(struct kexec_info 
> *UNUSED(info),
>  
>   end_phdr = _phdr[ehdr.e_phnum];
>  
> + /* Search for the real PAGE_OFFSET when KASLR memory randomization
> +  * is enabled */
> + if (get_kernel_sym("page_offset_base") != 0) {
> + for(phdr = ehdr.e_phdr; phdr != end_phdr; phdr++) {
> + if (phdr->p_type == PT_LOAD) {
> + vaddr = phdr->p_vaddr & pud_mask;
> + if (lowest_vaddr == 0 || lowest_vaddr > vaddr)
> + lowest_vaddr = vaddr;
> + }
> + }
> + if (lowest_vaddr != 0)
> + elf_info->page_offset = lowest_vaddr;
> + }
> +
>   /* Traverse through the Elf headers and find the region where
>* _stext symbol is located in. That's where kernel is mapped */
> - stext_sym = get_kernel_stext_sym();
> + stext_sym = get_kernel_sym("_stext");
>   for(phdr = ehdr.e_phdr; stext_sym && phdr != end_phdr; phdr++) {
>   if (phdr->p_type == PT_LOAD) {
>   unsigned long long saddr = phdr->p_vaddr;
> -- 
> 2.8.0.rc3.226.g39d4020
> 
> 
> ___
> kexec mailing list
> kexec@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/kexec

Thanks
Dave

___
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

Re: [PATCH v1] kexec/arch/i386: Add support for KASLR memory randomization

[Thomas Garnier]

On Wed, Aug 17, 2016 at 9:59 PM, Baoquan He  wrote:
> On 08/17/16 at 09:47am, Thomas Garnier wrote:
>> Multiple changes were made on KASLR (right now in linux-next). One of
>> them is randomizing the virtual address of the physical mapping, vmalloc
>> and vmemmap memory sections. It breaks kdump ability to read physical
>> memory.
>>
>> This change identifies if KASLR memories randomization is used by
>> checking if the page_offset_base variable exists. It search for the
>> correct PAGE_OFFSET value by looking at the loaded memory section and
>> find the lowest aligned on PUD (the randomization level).
>>
>> Related commits on linux-next:
>>  - 0483e1fa6e09d4948272680f691dccb1edb9677f: Base for randomization
>>  - 021182e52fe01c1f7b126f97fd6ba048dc4234fd: Enable for PAGE_OFFSET
>
> Seems above two commits have been inside Linus's tree, while vmemmap
> not yet.
>
>>
>> Signed-off-by: Thomas Garnier 
>> ---
>>  kexec/arch/i386/crashdump-x86.c | 29 ++---
>>  1 file changed, 22 insertions(+), 7 deletions(-)
>>
>> diff --git a/kexec/arch/i386/crashdump-x86.c 
>> b/kexec/arch/i386/crashdump-x86.c
>> index bbc0f35..ab833d4 100644
>> --- a/kexec/arch/i386/crashdump-x86.c
>> +++ b/kexec/arch/i386/crashdump-x86.c
>> @@ -102,11 +102,10 @@ static int get_kernel_paddr(struct kexec_info 
>> *UNUSED(info),
>>   return -1;
>>  }
>>
>> -/* Retrieve kernel _stext symbol virtual address from /proc/kallsyms */
>> -static unsigned long long get_kernel_stext_sym(void)
>> +/* Retrieve kernel symbol virtual address from /proc/kallsyms */
>> +static unsigned long long get_kernel_sym(const char *symbol)
>>  {
>>   const char *kallsyms = "/proc/kallsyms";
>> - const char *stext = "_stext";
>>   char sym[128];
>>   char line[128];
>>   FILE *fp;
>> @@ -122,13 +121,13 @@ static unsigned long long get_kernel_stext_sym(void)
>>   while(fgets(line, sizeof(line), fp) != NULL) {
>>   if (sscanf(line, "%Lx %c %s", , , sym) != 3)
>>   continue;
>> - if (strcmp(sym, stext) == 0) {
>> - dbgprintf("kernel symbol %s vaddr = %16llx\n", stext, 
>> vaddr);
>> + if (strcmp(sym, symbol) == 0) {
>> + dbgprintf("kernel symbol %s vaddr = %16llx\n", symbol, 
>> vaddr);
>>   return vaddr;
>>   }
>>   }
>>
>> - fprintf(stderr, "Cannot get kernel %s symbol address\n", stext);
>> + fprintf(stderr, "Cannot get kernel %s symbol address\n", symbol);
>>   return 0;
>>  }
>>
>> @@ -151,6 +150,8 @@ static int get_kernel_vaddr_and_size(struct kexec_info 
>> *UNUSED(info),
>>   off_t size;
>>   uint32_t elf_flags = 0;
>>   uint64_t stext_sym;
>> + const unsigned long long pud_mask = ~((1 << 30) - 1);
>> + unsigned long long vaddr, lowest_vaddr = 0;
>>
>>   if (elf_info->machine != EM_X86_64)
>>   return 0;
>> @@ -180,9 +181,23 @@ static int get_kernel_vaddr_and_size(struct kexec_info 
>> *UNUSED(info),
>>
>>   end_phdr = _phdr[ehdr.e_phnum];
>>
>> + /* Search for the real PAGE_OFFSET when KASLR memory randomization
>> +  * is enabled */
>
> Yeah, this is necessary. That would be great if it can be put into
> get_kernel_page_offset. But then it need parse kcore elf file again,
> seems no better way.
>

I agree.

Simon: Do you have any comments?

>> + if (get_kernel_sym("page_offset_base") != 0) {
>> + for(phdr = ehdr.e_phdr; phdr != end_phdr; phdr++) {
>> + if (phdr->p_type == PT_LOAD) {
>> + vaddr = phdr->p_vaddr & pud_mask;
>> + if (lowest_vaddr == 0 || lowest_vaddr > vaddr)
>> + lowest_vaddr = vaddr;
>> + }
>> + }
>> + if (lowest_vaddr != 0)
>> + elf_info->page_offset = lowest_vaddr;
>> + }
>> +
>>   /* Traverse through the Elf headers and find the region where
>>* _stext symbol is located in. That's where kernel is mapped */
>> - stext_sym = get_kernel_stext_sym();
>> + stext_sym = get_kernel_sym("_stext");
>>   for(phdr = ehdr.e_phdr; stext_sym && phdr != end_phdr; phdr++) {
>>   if (phdr->p_type == PT_LOAD) {
>>   unsigned long long saddr = phdr->p_vaddr;
>> --
>> 2.8.0.rc3.226.g39d4020
>>
>>
>> ___
>> kexec mailing list
>> kexec@lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/kexec

___
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

Re: [PATCH v1] kexec/arch/i386: Add support for KASLR memory randomization

[Baoquan He]

On 08/17/16 at 09:47am, Thomas Garnier wrote:
> Multiple changes were made on KASLR (right now in linux-next). One of
> them is randomizing the virtual address of the physical mapping, vmalloc
> and vmemmap memory sections. It breaks kdump ability to read physical
> memory.
> 
> This change identifies if KASLR memories randomization is used by
> checking if the page_offset_base variable exists. It search for the
> correct PAGE_OFFSET value by looking at the loaded memory section and
> find the lowest aligned on PUD (the randomization level).
> 
> Related commits on linux-next:
>  - 0483e1fa6e09d4948272680f691dccb1edb9677f: Base for randomization
>  - 021182e52fe01c1f7b126f97fd6ba048dc4234fd: Enable for PAGE_OFFSET

Seems above two commits have been inside Linus's tree, while vmemmap
not yet.

> 
> Signed-off-by: Thomas Garnier 
> ---
>  kexec/arch/i386/crashdump-x86.c | 29 ++---
>  1 file changed, 22 insertions(+), 7 deletions(-)
> 
> diff --git a/kexec/arch/i386/crashdump-x86.c b/kexec/arch/i386/crashdump-x86.c
> index bbc0f35..ab833d4 100644
> --- a/kexec/arch/i386/crashdump-x86.c
> +++ b/kexec/arch/i386/crashdump-x86.c
> @@ -102,11 +102,10 @@ static int get_kernel_paddr(struct kexec_info 
> *UNUSED(info),
>   return -1;
>  }
>  
> -/* Retrieve kernel _stext symbol virtual address from /proc/kallsyms */
> -static unsigned long long get_kernel_stext_sym(void)
> +/* Retrieve kernel symbol virtual address from /proc/kallsyms */
> +static unsigned long long get_kernel_sym(const char *symbol)
>  {
>   const char *kallsyms = "/proc/kallsyms";
> - const char *stext = "_stext";
>   char sym[128];
>   char line[128];
>   FILE *fp;
> @@ -122,13 +121,13 @@ static unsigned long long get_kernel_stext_sym(void)
>   while(fgets(line, sizeof(line), fp) != NULL) {
>   if (sscanf(line, "%Lx %c %s", , , sym) != 3)
>   continue;
> - if (strcmp(sym, stext) == 0) {
> - dbgprintf("kernel symbol %s vaddr = %16llx\n", stext, 
> vaddr);
> + if (strcmp(sym, symbol) == 0) {
> + dbgprintf("kernel symbol %s vaddr = %16llx\n", symbol, 
> vaddr);
>   return vaddr;
>   }
>   }
>  
> - fprintf(stderr, "Cannot get kernel %s symbol address\n", stext);
> + fprintf(stderr, "Cannot get kernel %s symbol address\n", symbol);
>   return 0;
>  }
>  
> @@ -151,6 +150,8 @@ static int get_kernel_vaddr_and_size(struct kexec_info 
> *UNUSED(info),
>   off_t size;
>   uint32_t elf_flags = 0;
>   uint64_t stext_sym;
> + const unsigned long long pud_mask = ~((1 << 30) - 1);
> + unsigned long long vaddr, lowest_vaddr = 0;
>  
>   if (elf_info->machine != EM_X86_64)
>   return 0;
> @@ -180,9 +181,23 @@ static int get_kernel_vaddr_and_size(struct kexec_info 
> *UNUSED(info),
>  
>   end_phdr = _phdr[ehdr.e_phnum];
>  
> + /* Search for the real PAGE_OFFSET when KASLR memory randomization
> +  * is enabled */

Yeah, this is necessary. That would be great if it can be put into
get_kernel_page_offset. But then it need parse kcore elf file again,
seems no better way.

> + if (get_kernel_sym("page_offset_base") != 0) {
> + for(phdr = ehdr.e_phdr; phdr != end_phdr; phdr++) {
> + if (phdr->p_type == PT_LOAD) {
> + vaddr = phdr->p_vaddr & pud_mask;
> + if (lowest_vaddr == 0 || lowest_vaddr > vaddr)
> + lowest_vaddr = vaddr;
> + }
> + }
> + if (lowest_vaddr != 0)
> + elf_info->page_offset = lowest_vaddr;
> + }
> +
>   /* Traverse through the Elf headers and find the region where
>* _stext symbol is located in. That's where kernel is mapped */
> - stext_sym = get_kernel_stext_sym();
> + stext_sym = get_kernel_sym("_stext");
>   for(phdr = ehdr.e_phdr; stext_sym && phdr != end_phdr; phdr++) {
>   if (phdr->p_type == PT_LOAD) {
>   unsigned long long saddr = phdr->p_vaddr;
> -- 
> 2.8.0.rc3.226.g39d4020
> 
> 
> ___
> kexec mailing list
> kexec@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/kexec

___
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec