[LTP] [PATCH] mmap04.c: Avoid vma merging

[Avinesh Kumar]

Hi Martin,

On Wednesday, January 24, 2024 12:56:58 PM CET Martin Doucha wrote:
> Hi,
> some comments below.
> 
> On 23. 01. 24 17:55, Avinesh Kumar wrote:
> > We hit a scenario where new mapping was merged with existing mapping of
> > same permission and the return address from the mmap was hidden in the
> > merged mapping in /proc/self/maps, causing the test to fail.
> > To avoid this, we first  create a 2-page mapping with the different
> > permissions, and then remap the 2nd page with the perms being tested.
> > 
> > Signed-off-by: Avinesh Kumar <akumar@suse.de>
> > Reported-by: Martin Doucha <mdoucha@suse.cz>
> > ---
> > 
> >   testcases/kernel/syscalls/mmap/mmap04.c | 49 +++++++++++++++----------
> >   1 file changed, 30 insertions(+), 19 deletions(-)
> > 
> > diff --git a/testcases/kernel/syscalls/mmap/mmap04.c
> > b/testcases/kernel/syscalls/mmap/mmap04.c index f6f4f7c98..f0f87b7f5
> > 100644
> > --- a/testcases/kernel/syscalls/mmap/mmap04.c
> > +++ b/testcases/kernel/syscalls/mmap/mmap04.c
> > @@ -17,28 +17,28 @@
> > 
> >   #include "tst_test.h"
> >   #include <stdio.h>
> > 
> > -#define MMAPSIZE 1024
> > -static char *addr;
> > +static char *addr1;
> > +static char *addr2;
> > 
> >   static struct tcase {
> >   
> >   	int prot;
> >   	int flags;
> >   	char *exp_perms;
> >   
> >   } tcases[] = {
> > 
> > -	{PROT_NONE, MAP_ANONYMOUS | MAP_PRIVATE, "---p"},
> > -	{PROT_NONE, MAP_ANONYMOUS | MAP_SHARED, "---s"},
> > -	{PROT_READ, MAP_ANONYMOUS | MAP_PRIVATE, "r--p"},
> > -	{PROT_READ, MAP_ANONYMOUS | MAP_SHARED, "r--s"},
> > -	{PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, "-w-p"},
> > -	{PROT_WRITE, MAP_ANONYMOUS | MAP_SHARED, "-w-s"},
> > -	{PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, "rw-p"},
> > -	{PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_SHARED, "rw-s"},
> > -	{PROT_READ | PROT_EXEC, MAP_ANONYMOUS | MAP_PRIVATE, "r-xp"},
> > -	{PROT_READ | PROT_EXEC, MAP_ANONYMOUS | MAP_SHARED, "r-xs"},
> > -	{PROT_WRITE | PROT_EXEC, MAP_ANONYMOUS | MAP_PRIVATE, "-wxp"},
> > -	{PROT_WRITE | PROT_EXEC, MAP_ANONYMOUS | MAP_SHARED, "-wxs"},
> > -	{PROT_READ | PROT_WRITE | PROT_EXEC, MAP_ANONYMOUS | MAP_PRIVATE,
> > "rwxp"}, -	{PROT_READ | PROT_WRITE | PROT_EXEC, MAP_ANONYMOUS |
> > MAP_SHARED, "rwxs"} +	{PROT_NONE, MAP_ANONYMOUS | MAP_PRIVATE |
> > MAP_FIXED, "---p"},
> > +	{PROT_NONE, MAP_ANONYMOUS | MAP_SHARED | MAP_FIXED, "---s"},
> > +	{PROT_READ, MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, "r--p"},
> > +	{PROT_READ, MAP_ANONYMOUS | MAP_SHARED | MAP_FIXED, "r--s"},
> > +	{PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, "-w-p"},
> > +	{PROT_WRITE, MAP_ANONYMOUS | MAP_SHARED | MAP_FIXED, "-w-s"},
> > +	{PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED,
> > "rw-p"}, +	{PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_SHARED |
> > MAP_FIXED, "rw-s"}, +	{PROT_READ | PROT_EXEC, MAP_ANONYMOUS | MAP_PRIVATE
> > | MAP_FIXED, "r-xp"}, +	{PROT_READ | PROT_EXEC, MAP_ANONYMOUS |
> > MAP_SHARED | MAP_FIXED, "r-xs"}, +	{PROT_WRITE | PROT_EXEC, 
MAP_ANONYMOUS
> > | MAP_PRIVATE | MAP_FIXED, "-wxp"}, +	{PROT_WRITE | PROT_EXEC,
> > MAP_ANONYMOUS | MAP_SHARED | MAP_FIXED, "-wxs"}, +	{PROT_READ |
> > PROT_WRITE | PROT_EXEC, MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, "rwxp"},
> > +	{PROT_READ | PROT_WRITE | PROT_EXEC, MAP_ANONYMOUS | MAP_SHARED |
> > MAP_FIXED, "rwxs"}
> The MAP_FIXED flag doesn't belong in the testcases, it should be added
> in the mmap() call instead: SAFE_MMAP(..., tc->flags | MAP_FIXED, ...);
> It's an implementation detail not related to the testcases themselves.
> You don't want to rewrite all the test cases again if we decide to not
> use MAP_FIXED for whatever reason in the future.
> 
Thank you for review and all the corrections/suggestions. I have send the 
updated patch.

> >   };
> >   
> >   static void run(unsigned int i)
> > 
> > @@ -47,10 +47,21 @@ static void run(unsigned int i)
> > 
> >   	char addr_str[20];
> >   	char perms[8];
> >   	char fmt[1024];
> > 
> > +	unsigned int pagesize;
> > 
> > -	addr = SAFE_MMAP(NULL, MMAPSIZE, tc->prot, tc->flags, -1, 0);
> > +	pagesize = SAFE_SYSCONF(_SC_PAGESIZE);
> > 
> > -	sprintf(addr_str, "%" PRIxPTR, (uintptr_t)addr);
> > +	/* To avoid new mapping getting merged with existing mappings, we 
first
> > +	   create a 2-page mapping with the different permissions, and then
> > remap
> > +	   the 2nd page with the perms being tested. */
> > +	if ((tc->prot == PROT_NONE) && (tc->flags == (MAP_ANONYMOUS |
> > MAP_PRIVATE | MAP_FIXED))) +		addr1 = SAFE_MMAP(NULL, 
pagesize * 2,
> > PROT_READ, MAP_ANONYMOUS | MAP_SHARED, -1, 0); +	else
> > +		addr1 = SAFE_MMAP(NULL, pagesize * 2, PROT_NONE, 
MAP_ANONYMOUS |
> > MAP_PRIVATE, -1, 0);
> This would be cleaner (just invert the shared/private flag):
> int flags = (tc->flags & MAP_PRIVATE) ? MAP_SHARED : MAP_PRIVATE;
> addr1 = SAFE_MMAP(NULL, pagesize * 2, PROT_NONE, MAP_ANONYMOUS | flags,
> -1, 0);
> 
> > +
> > +	addr2 = SAFE_MMAP(addr1 + pagesize, pagesize, tc->prot, tc->flags, 
-1,
> > 0); +
> > +	sprintf(addr_str, "%" PRIxPTR, (uintptr_t)addr2);
> 
> Why not merge the two sprintf()s into one?
> sprintf(fmt, "%" PRIxPTR "-%%*x %%s", (uintptr_t)addr2);
> 
> >   	sprintf(fmt, "%s-%%*x %%s", addr_str);
> >   	SAFE_FILE_LINES_SCANF("/proc/self/maps", fmt, perms);
> > 
> > @@ -61,7 +72,7 @@ static void run(unsigned int i)
> > 
> >   						tc-
>exp_perms, perms);
> >   	
> >   	}
> > 
> > -	SAFE_MUNMAP(addr, MMAPSIZE);
> > +	SAFE_MUNMAP(addr1, pagesize * 2);
> > 
> >   }
> >   
> >   static struct tst_test test = {