[LTP] [PATCH] mmap1: fix race between map_write_unmap and read_mem

[LTP] [PATCH] mmap1: fix race between map_write_unmap and read_mem

[Dmitry Guryanov]

On one of test runs it failed with follwing log:

mmap1       0  INFO  :  created thread[1093683520]
mmap1       0  INFO  :  pid[6430]: map, change contents, unmap files 1000 times
mmap1       0  INFO  :  created thread[1109268800]
mmap1       0  INFO  :  pid[6430] - read contents of memory (nil) 1000 times
mmap1       0  INFO  :  page fault occurred at (nil)
mmap1       0  INFO  :  page fault occurred at (nil)
....................
mmap1       0  INFO  :  page fault occurred at (nil)
caught unexpected signal - 11 --- exiting

This can happen, if thread map_write_unmap haven't mapped file yet, but
thread read_mem already began to read from it (it reads from map_address, which
will be NULL in this case).

Let's wait, while thread map_write_unmap actually maps file, in read_mem thread
and zero map_address each time map_write_unmap unmaps this memory.
---
 testcases/kernel/mem/mtest06/mmap1.c |    5 +++++
 1 files changed, 5 insertions(+), 0 deletions(-)

diff --git a/testcases/kernel/mem/mtest06/mmap1.c b/testcases/kernel/mem/mtest06/mmap1.c
index 47ab49a..28d9e5d 100644
--- a/testcases/kernel/mem/mtest06/mmap1.c
+++ b/testcases/kernel/mem/mtest06/mmap1.c
@@ -331,6 +331,7 @@ map_write_unmap(void *args)	/* file descriptor of the file to be mapped.  */
             exit_val = MWU_FAIL;
             pthread_exit((void *)exit_val);
         }
+        map_address = NULL;
     }
     exit_val = M_SUCCESS;
     pthread_exit((void *)exit_val);
@@ -358,6 +359,10 @@ read_mem(void *args)		/* number of reads performed		      */
     long 	*rmargs = args;	/* local pointer to the arguments	      */
     long exit_val = 0;  /* pthread exit value			      */
 
+    /* wait for first thread to map file */
+    while(!map_address)
+        sched_yield();
+
     tst_resm(TINFO, "pid[%d] - read contents of memory %p %ld times",
            getpid(), map_address, rmargs[2]);
     if (verbose_print)
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel® Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] mmap1: fix race between map_write_unmap and read_mem

[avagin]

Hi, all.

We have seen on this test case in more detail and find some race conditions.

1. the first race between setting & reading & checking & printing value 
of map_address

This race is reproduced if a reader thread is started before a 
map_write_unmap thread. I added logging and got follow output:

caught unexpected signal - 11 info->si_addr=(nil) 
map_address=0x2aaaaaaac000--- exiting

[thread]	:	[operation]
read_mem	:	if (strncmp((char *)map_address, "a", 1) != 0)
map_write_unmap	:	map_address = mmap(...)
read_mem	:	info->si_addr == map_address

2. the second race between memcpy and a check of memory content.
[thread]	:	[operation]
map_write_unmap	:	map_address = mmap(...)
read_mem	:	if (strncmp((char *)map_address, "a", 1) != 0)
map_write_unmap	:	memset(map_address, 'a', mwuargs[1]);


On 03/23/2010 10:03 PM, Dmitry Guryanov wrote:
> On one of test runs it failed with follwing log:
>
> mmap1       0  INFO  :  created thread[1093683520]
> mmap1       0  INFO  :  pid[6430]: map, change contents, unmap files 1000 times
> mmap1       0  INFO  :  created thread[1109268800]
> mmap1       0  INFO  :  pid[6430] - read contents of memory (nil) 1000 times
> mmap1       0  INFO  :  page fault occurred at (nil)
> mmap1       0  INFO  :  page fault occurred at (nil)
> ....................
> mmap1       0  INFO  :  page fault occurred at (nil)
> caught unexpected signal - 11 --- exiting
>
> This can happen, if thread map_write_unmap haven't mapped file yet, but
> thread read_mem already began to read from it (it reads from map_address, which
> will be NULL in this case).

This testcase tries to handle SIGSEGV:
in reader thread, it does setjmp(), and in signal handler it does longjmp().

>
> Let's wait, while thread map_write_unmap actually maps file, in read_mem thread
> and zero map_address each time map_write_unmap unmaps this memory.
> ---
>   testcases/kernel/mem/mtest06/mmap1.c |    5 +++++
>   1 files changed, 5 insertions(+), 0 deletions(-)
>
> diff --git a/testcases/kernel/mem/mtest06/mmap1.c b/testcases/kernel/mem/mtest06/mmap1.c
> index 47ab49a..28d9e5d 100644
> --- a/testcases/kernel/mem/mtest06/mmap1.c
> +++ b/testcases/kernel/mem/mtest06/mmap1.c
> @@ -331,6 +331,7 @@ map_write_unmap(void *args)	/* file descriptor of the file to be mapped.  */
>               exit_val = MWU_FAIL;
>               pthread_exit((void *)exit_val);
>           }
> +        map_address = NULL;
>       }
>       exit_val = M_SUCCESS;
>       pthread_exit((void *)exit_val);
> @@ -358,6 +359,10 @@ read_mem(void *args)		/* number of reads performed		      */
>       long 	*rmargs = args;	/* local pointer to the arguments	      */
>       long exit_val = 0;  /* pthread exit value			      */
>
> +    /* wait for first thread to map file */
> +    while(!map_address)
> +        sched_yield();
> +
>       tst_resm(TINFO, "pid[%d] - read contents of memory %p %ld times",
>              getpid(), map_address, rmargs[2]);
>       if (verbose_print)


------------------------------------------------------------------------------
Download Intel® Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list