linux-user: use default mmap_min_addr 65536 when /proc/sys/vm/mmap_min_addr cannot be read
PATCH 1/1 linux-user: use default mmap_min_addr 65536 when /proc/sys/vm/mmap_min_addr cannot be read
* 65536 is default at least for ubuntu and fedora.
* after http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=0e1a6ef2dea88101b056b6d9984f3325c5efced3
it's not readable by normal user on many systems
* there is patch to enable read-only access again
http://git.kernel.org/?p=linux/kernel/git/jmorris/security-testing-2.6.git;a=commitdiff;h=822cceec7248013821d655545ea45d1c6a9d15b3
but that's not applied in many distributions
* please consider this workaround
--- linux-user/main.c | 7 +++++++ 1 files changed, 7 insertions(+), 0 deletions(-)
linux-user/main.c +7 -0
@@ -2891,6 +2891,7 @@ int main(int argc, char **argv, char **envp)* When loading the ELF image to determine whether guest_base * is needed. It is also used in mmap_find_vma. */ +#define MMAP_MIN_ADDR_DEFAULT 65536 {FILE *fp;
@@ -2899,8 +2900,14 @@ int main(int argc, char **argv, char **envp)if (fscanf (fp,"%lu" , &tmp) ==1 ) { mmap_min_addr = tmp;qemu_log ("host mmap_min_addr=0x%lx\n" , mmap_min_addr); + }else { +qemu_log ("cannot read value from /proc/sys/vm/mmap_min_addr, assuming %d\n" ,MMAP_MIN_ADDR_DEFAULT ); + mmap_min_addr =MMAP_MIN_ADDR_DEFAULT ; }fclose (fp); + }else { +qemu_log ("cannot open /proc/sys/vm/mmap_min_addr for reading, assuming %d\n" ,MMAP_MIN_ADDR_DEFAULT ); + mmap_min_addr =MMAP_MIN_ADDR_DEFAULT ; } }
Signature · 1 line
1.7.1
Quoted 4 lines · Martin Jansa
linux-user/main.c +0
-0@@ -2899,8 +2900,14 @@ int main(int argc, char **argv, char **envp)if (fscanf (fp,"%lu" , &tmp) ==1 ) { mmap_min_addr = tmp;qemu_log ("host mmap_min_addr=0x%lx\n" , mmap_min_addr);
linux-user/main.c int main(int argc, char **argv, char **envp) · line 2912
Quoted 10 lines
linux-user/main.c +6
-0+ }else { +qemu_log ("cannot read value from /proc/sys/vm/mmap_min_addr, assuming %d\n" ,MMAP_MIN_ADDR_DEFAULT ); + mmap_min_addr =MMAP_MIN_ADDR_DEFAULT ; }fclose (fp); + }else { +qemu_log ("cannot open /proc/sys/vm/mmap_min_addr for reading, assuming %d\n" ,MMAP_MIN_ADDR_DEFAULT ); + mmap_min_addr =MMAP_MIN_ADDR_DEFAULT ; } }
Perhaps you can combine these two else clauses?
r~
Quoted 17 lines · Richard Henderson
Quoted 4 lines · Martin Jansa
linux-user/main.c +0
-0@@ -2899,8 +2900,14 @@ int main(int argc, char **argv, char **envp)if (fscanf (fp,"%lu" , &tmp) ==1 ) { mmap_min_addr = tmp;qemu_log ("host mmap_min_addr=0x%lx\n" , mmap_min_addr);linux-user/main.c int main(int argc, char **argv, char **envp) · line 2912
Quoted 10 lines
linux-user/main.c +6
-0+ }else { +qemu_log ("cannot read value from /proc/sys/vm/mmap_min_addr, assuming %d\n" ,MMAP_MIN_ADDR_DEFAULT ); + mmap_min_addr =MMAP_MIN_ADDR_DEFAULT ; }fclose (fp); + }else { +qemu_log ("cannot open /proc/sys/vm/mmap_min_addr for reading, assuming %d\n" ,MMAP_MIN_ADDR_DEFAULT ); + mmap_min_addr =MMAP_MIN_ADDR_DEFAULT ; } }Perhaps you can combine these two else clauses?
Sure it would be possible, but while trying to find why qemu doesn't work on someone box but works on mine I wanted to distinguish between cannot open and cannot read.
Also if I combine it in test ie initialize mmap_min_addr with -1, and then check if it changed after read (because 0 is also valid value), or moving fopen and fscanf to one condition, then it gets IMHO as complicated as those 2 else clauses.
BTW: I noticed that qemu-arm works ok when mmap_min_addr is low enough (and doesn't have to be zero).
my box had mmap_min_addr 4096 and qemu-arm worked fine even without reading it
and the other boxes had mmap_min_addr = 64K and failed later to mmap.
Regards,
Signature · 3 lines
uin:136542059 jid:[email protected] Jansa Martin sip:[email protected] JaMa
Quoted 5 lines · Martin Jansa
BTW: I noticed that qemu-arm works ok when
mmap_min_addris low enough (and doesn't have to be zero).my box had
mmap_min_addr4096 and qemu-arm worked fine even without reading it
Paul Brook's c581deda322080e8beb88b2e468d4af54454e4b3 had the side effect of working around mmap_min_addr without having to read it. Of course, if we *can* read mmap_min_addr, his patch will do less work.
r~