diff mbox series

[for-8.1,v10,04/14] linux-user: Use MAP_FIXED_NOREPLACE for initial image mmap

Message ID 20230807163705.9848-5-richard.henderson@linaro.org
State Superseded
Headers show
Series linux-user: image mapping fixes | expand

Commit Message

Richard Henderson Aug. 7, 2023, 4:36 p.m. UTC
Use this as extra protection for the guest mapping over
any qemu host mappings.

Tested-by: Helge Deller <deller@gmx.de>
Reviewed-by: Helge Deller <deller@gmx.de>
Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 linux-user/elfload.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

Comments

Alex Bennée Aug. 8, 2023, 9:43 a.m. UTC | #1
Richard Henderson <richard.henderson@linaro.org> writes:

> Use this as extra protection for the guest mapping over
> any qemu host mappings.
>
> Tested-by: Helge Deller <deller@gmx.de>
> Reviewed-by: Helge Deller <deller@gmx.de>
> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  linux-user/elfload.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/linux-user/elfload.c b/linux-user/elfload.c
> index 36e4026f05..1b4bb2d5af 100644
> --- a/linux-user/elfload.c
> +++ b/linux-user/elfload.c
> @@ -3147,8 +3147,11 @@ static void load_elf_image(const char *image_name, int image_fd,
>      /*
>       * Reserve address space for all of this.
>       *
> -     * In the case of ET_EXEC, we supply MAP_FIXED so that we get
> -     * exactly the address range that is required.
> +     * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
> +     * exactly the address range that is required.  Without reserved_va,
> +     * the guest address space is not isolated.  We have attempted to avoid
> +     * conflict with the host program itself via probe_guest_base, but using
> +     * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
>       *
>       * Otherwise this is ET_DYN, and we are searching for a location
>       * that can hold the memory space required.  If the image is
> @@ -3160,7 +3163,7 @@ static void load_elf_image(const char *image_name, int image_fd,
>       */
>      load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
>                              MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
> -                            (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
> +                            (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
>                              -1, 0);

We should probably also check the result == load_addr for the places
where MAP_FIXED_NOREPLACE isn't supported as we have this in osdep.h:

  #ifndef MAP_FIXED_NOREPLACE
  #define MAP_FIXED_NOREPLACE 0
  #endif

See 2667e069e7 (linux-user: don't use MAP_FIXED in pgd_find_hole_fallback)

>      if (load_addr == -1) {
>          goto exit_mmap;
Akihiko Odaki Aug. 8, 2023, 11:57 a.m. UTC | #2
On 2023/08/08 18:43, Alex Bennée wrote:
> 
> Richard Henderson <richard.henderson@linaro.org> writes:
> 
>> Use this as extra protection for the guest mapping over
>> any qemu host mappings.
>>
>> Tested-by: Helge Deller <deller@gmx.de>
>> Reviewed-by: Helge Deller <deller@gmx.de>
>> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>> ---
>>   linux-user/elfload.c | 9 ++++++---
>>   1 file changed, 6 insertions(+), 3 deletions(-)
>>
>> diff --git a/linux-user/elfload.c b/linux-user/elfload.c
>> index 36e4026f05..1b4bb2d5af 100644
>> --- a/linux-user/elfload.c
>> +++ b/linux-user/elfload.c
>> @@ -3147,8 +3147,11 @@ static void load_elf_image(const char *image_name, int image_fd,
>>       /*
>>        * Reserve address space for all of this.
>>        *
>> -     * In the case of ET_EXEC, we supply MAP_FIXED so that we get
>> -     * exactly the address range that is required.
>> +     * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
>> +     * exactly the address range that is required.  Without reserved_va,
>> +     * the guest address space is not isolated.  We have attempted to avoid
>> +     * conflict with the host program itself via probe_guest_base, but using
>> +     * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
>>        *
>>        * Otherwise this is ET_DYN, and we are searching for a location
>>        * that can hold the memory space required.  If the image is
>> @@ -3160,7 +3163,7 @@ static void load_elf_image(const char *image_name, int image_fd,
>>        */
>>       load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
>>                               MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
>> -                            (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
>> +                            (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
>>                               -1, 0);
> 
> We should probably also check the result == load_addr for the places
> where MAP_FIXED_NOREPLACE isn't supported as we have this in osdep.h:
> 
>    #ifndef MAP_FIXED_NOREPLACE
>    #define MAP_FIXED_NOREPLACE 0
>    #endif
> 
> See 2667e069e7 (linux-user: don't use MAP_FIXED in pgd_find_hole_fallback)

It assumes target_mmap() emulates MAP_FIXED_NOREPLACE when the host does 
not support it as commit e69e032d1a ("linux-user: Use 
MAP_FIXED_NOREPLACE for do_brk()") already does, but defining 
MAP_FIXED_NOREPLACE zero breaks such emulation. I wrote a fix:
https://patchew.org/QEMU/20230808115242.73025-1-akihiko.odaki@daynix.com/

> 
>>       if (load_addr == -1) {
>>           goto exit_mmap;
> 
>
Alex Bennée Aug. 8, 2023, 1:48 p.m. UTC | #3
Akihiko Odaki <akihiko.odaki@daynix.com> writes:

> On 2023/08/08 18:43, Alex Bennée wrote:
>> Richard Henderson <richard.henderson@linaro.org> writes:
>> 
>>> Use this as extra protection for the guest mapping over
>>> any qemu host mappings.
>>>
>>> Tested-by: Helge Deller <deller@gmx.de>
>>> Reviewed-by: Helge Deller <deller@gmx.de>
>>> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
>>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>>> ---
>>>   linux-user/elfload.c | 9 ++++++---
>>>   1 file changed, 6 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/linux-user/elfload.c b/linux-user/elfload.c
>>> index 36e4026f05..1b4bb2d5af 100644
>>> --- a/linux-user/elfload.c
>>> +++ b/linux-user/elfload.c
>>> @@ -3147,8 +3147,11 @@ static void load_elf_image(const char *image_name, int image_fd,
>>>       /*
>>>        * Reserve address space for all of this.
>>>        *
>>> -     * In the case of ET_EXEC, we supply MAP_FIXED so that we get
>>> -     * exactly the address range that is required.
>>> +     * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
>>> +     * exactly the address range that is required.  Without reserved_va,
>>> +     * the guest address space is not isolated.  We have attempted to avoid
>>> +     * conflict with the host program itself via probe_guest_base, but using
>>> +     * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
>>>        *
>>>        * Otherwise this is ET_DYN, and we are searching for a location
>>>        * that can hold the memory space required.  If the image is
>>> @@ -3160,7 +3163,7 @@ static void load_elf_image(const char *image_name, int image_fd,
>>>        */
>>>       load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
>>>                               MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
>>> -                            (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
>>> +                            (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
>>>                               -1, 0);
>> We should probably also check the result == load_addr for the places
>> where MAP_FIXED_NOREPLACE isn't supported as we have this in osdep.h:
>>    #ifndef MAP_FIXED_NOREPLACE
>>    #define MAP_FIXED_NOREPLACE 0
>>    #endif
>> See 2667e069e7 (linux-user: don't use MAP_FIXED in
>> pgd_find_hole_fallback)
>
> It assumes target_mmap() emulates MAP_FIXED_NOREPLACE when the host
> does not support it as commit e69e032d1a ("linux-user: Use
> MAP_FIXED_NOREPLACE for do_brk()") already does, but defining
> MAP_FIXED_NOREPLACE zero breaks such emulation. I wrote a fix:
> https://patchew.org/QEMU/20230808115242.73025-1-akihiko.odaki@daynix.com/

Hmm doesn't that push the problem to real mmap() calls to a host system
that doesn't support MAP_FIXED_NOREPLACE?

I wonder if we need an internal flag rather than overloading the host
flags?

>
>> 
>>>       if (load_addr == -1) {
>>>           goto exit_mmap;
>>
Akihiko Odaki Aug. 8, 2023, 2:08 p.m. UTC | #4
On 2023/08/08 22:48, Alex Bennée wrote:
> 
> Akihiko Odaki <akihiko.odaki@daynix.com> writes:
> 
>> On 2023/08/08 18:43, Alex Bennée wrote:
>>> Richard Henderson <richard.henderson@linaro.org> writes:
>>>
>>>> Use this as extra protection for the guest mapping over
>>>> any qemu host mappings.
>>>>
>>>> Tested-by: Helge Deller <deller@gmx.de>
>>>> Reviewed-by: Helge Deller <deller@gmx.de>
>>>> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
>>>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>>>> ---
>>>>    linux-user/elfload.c | 9 ++++++---
>>>>    1 file changed, 6 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/linux-user/elfload.c b/linux-user/elfload.c
>>>> index 36e4026f05..1b4bb2d5af 100644
>>>> --- a/linux-user/elfload.c
>>>> +++ b/linux-user/elfload.c
>>>> @@ -3147,8 +3147,11 @@ static void load_elf_image(const char *image_name, int image_fd,
>>>>        /*
>>>>         * Reserve address space for all of this.
>>>>         *
>>>> -     * In the case of ET_EXEC, we supply MAP_FIXED so that we get
>>>> -     * exactly the address range that is required.
>>>> +     * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
>>>> +     * exactly the address range that is required.  Without reserved_va,
>>>> +     * the guest address space is not isolated.  We have attempted to avoid
>>>> +     * conflict with the host program itself via probe_guest_base, but using
>>>> +     * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
>>>>         *
>>>>         * Otherwise this is ET_DYN, and we are searching for a location
>>>>         * that can hold the memory space required.  If the image is
>>>> @@ -3160,7 +3163,7 @@ static void load_elf_image(const char *image_name, int image_fd,
>>>>         */
>>>>        load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
>>>>                                MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
>>>> -                            (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
>>>> +                            (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
>>>>                                -1, 0);
>>> We should probably also check the result == load_addr for the places
>>> where MAP_FIXED_NOREPLACE isn't supported as we have this in osdep.h:
>>>     #ifndef MAP_FIXED_NOREPLACE
>>>     #define MAP_FIXED_NOREPLACE 0
>>>     #endif
>>> See 2667e069e7 (linux-user: don't use MAP_FIXED in
>>> pgd_find_hole_fallback)
>>
>> It assumes target_mmap() emulates MAP_FIXED_NOREPLACE when the host
>> does not support it as commit e69e032d1a ("linux-user: Use
>> MAP_FIXED_NOREPLACE for do_brk()") already does, but defining
>> MAP_FIXED_NOREPLACE zero breaks such emulation. I wrote a fix:
>> https://patchew.org/QEMU/20230808115242.73025-1-akihiko.odaki@daynix.com/
> 
> Hmm doesn't that push the problem to real mmap() calls to a host system
> that doesn't support MAP_FIXED_NOREPLACE?

That can happen even without that patch if you run QEMU built with a new 
libc on old Linux so we should have prepared for that kind of situation.

The man page also says:
 > Note that older kernels which do not recognize the MAP_FIXED_NOREPLACE
 > flag will typically (upon detecting a collision with a preexisting
 > mapping) fall back to a “non-MAP_FIXED” type of behavior: they will
 > return an address that is different from the requested address.
 > Therefore, backward-compatible software should check the returned
 > address against the requested address.
https://man7.org/linux/man-pages/man2/mmap.2.html

It basically means MAP_FIXED_NOREPLACE has no effect on a host that 
doesn't support it, and the existing code checking the returned address 
should continue to work.

> 
> I wonder if we need an internal flag rather than overloading the host
> flags?


> 
>>
>>>
>>>>        if (load_addr == -1) {
>>>>            goto exit_mmap;
>>>
> 
>
Alex Bennée Aug. 8, 2023, 2:20 p.m. UTC | #5
Akihiko Odaki <akihiko.odaki@daynix.com> writes:

> On 2023/08/08 22:48, Alex Bennée wrote:
>> Akihiko Odaki <akihiko.odaki@daynix.com> writes:
>> 
>>> On 2023/08/08 18:43, Alex Bennée wrote:
>>>> Richard Henderson <richard.henderson@linaro.org> writes:
>>>>
>>>>> Use this as extra protection for the guest mapping over
>>>>> any qemu host mappings.
>>>>>
>>>>> Tested-by: Helge Deller <deller@gmx.de>
>>>>> Reviewed-by: Helge Deller <deller@gmx.de>
>>>>> Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
>>>>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>>>>> ---
>>>>>    linux-user/elfload.c | 9 ++++++---
>>>>>    1 file changed, 6 insertions(+), 3 deletions(-)
>>>>>
>>>>> diff --git a/linux-user/elfload.c b/linux-user/elfload.c
>>>>> index 36e4026f05..1b4bb2d5af 100644
>>>>> --- a/linux-user/elfload.c
>>>>> +++ b/linux-user/elfload.c
>>>>> @@ -3147,8 +3147,11 @@ static void load_elf_image(const char *image_name, int image_fd,
>>>>>        /*
>>>>>         * Reserve address space for all of this.
>>>>>         *
>>>>> -     * In the case of ET_EXEC, we supply MAP_FIXED so that we get
>>>>> -     * exactly the address range that is required.
>>>>> +     * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
>>>>> +     * exactly the address range that is required.  Without reserved_va,
>>>>> +     * the guest address space is not isolated.  We have attempted to avoid
>>>>> +     * conflict with the host program itself via probe_guest_base, but using
>>>>> +     * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
>>>>>         *
>>>>>         * Otherwise this is ET_DYN, and we are searching for a location
>>>>>         * that can hold the memory space required.  If the image is
>>>>> @@ -3160,7 +3163,7 @@ static void load_elf_image(const char *image_name, int image_fd,
>>>>>         */
>>>>>        load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
>>>>>                                MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
>>>>> -                            (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
>>>>> +                            (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
>>>>>                                -1, 0);
>>>> We should probably also check the result == load_addr for the places
>>>> where MAP_FIXED_NOREPLACE isn't supported as we have this in osdep.h:
>>>>     #ifndef MAP_FIXED_NOREPLACE
>>>>     #define MAP_FIXED_NOREPLACE 0
>>>>     #endif
>>>> See 2667e069e7 (linux-user: don't use MAP_FIXED in
>>>> pgd_find_hole_fallback)
>>>
>>> It assumes target_mmap() emulates MAP_FIXED_NOREPLACE when the host
>>> does not support it as commit e69e032d1a ("linux-user: Use
>>> MAP_FIXED_NOREPLACE for do_brk()") already does, but defining
>>> MAP_FIXED_NOREPLACE zero breaks such emulation. I wrote a fix:
>>> https://patchew.org/QEMU/20230808115242.73025-1-akihiko.odaki@daynix.com/
>> Hmm doesn't that push the problem to real mmap() calls to a host
>> system
>> that doesn't support MAP_FIXED_NOREPLACE?
>
> That can happen even without that patch if you run QEMU built with a
> new libc on old Linux so we should have prepared for that kind of
> situation.
>
> The man page also says:
>> Note that older kernels which do not recognize the MAP_FIXED_NOREPLACE
>> flag will typically (upon detecting a collision with a preexisting
>> mapping) fall back to a “non-MAP_FIXED” type of behavior: they will
>> return an address that is different from the requested address.
>> Therefore, backward-compatible software should check the returned
>> address against the requested address.
> https://man7.org/linux/man-pages/man2/mmap.2.html
>
> It basically means MAP_FIXED_NOREPLACE has no effect on a host that
> doesn't support it, and the existing code checking the returned
> address should continue to work.

OK - as long as it doesn't barf on unknown flags.
diff mbox series

Patch

diff --git a/linux-user/elfload.c b/linux-user/elfload.c
index 36e4026f05..1b4bb2d5af 100644
--- a/linux-user/elfload.c
+++ b/linux-user/elfload.c
@@ -3147,8 +3147,11 @@  static void load_elf_image(const char *image_name, int image_fd,
     /*
      * Reserve address space for all of this.
      *
-     * In the case of ET_EXEC, we supply MAP_FIXED so that we get
-     * exactly the address range that is required.
+     * In the case of ET_EXEC, we supply MAP_FIXED_NOREPLACE so that we get
+     * exactly the address range that is required.  Without reserved_va,
+     * the guest address space is not isolated.  We have attempted to avoid
+     * conflict with the host program itself via probe_guest_base, but using
+     * MAP_FIXED_NOREPLACE instead of MAP_FIXED provides an extra check.
      *
      * Otherwise this is ET_DYN, and we are searching for a location
      * that can hold the memory space required.  If the image is
@@ -3160,7 +3163,7 @@  static void load_elf_image(const char *image_name, int image_fd,
      */
     load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
                             MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
-                            (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
+                            (ehdr->e_type == ET_EXEC ? MAP_FIXED_NOREPLACE : 0),
                             -1, 0);
     if (load_addr == -1) {
         goto exit_mmap;