crypto/af_alg02: fixed read() being stuck
PATCH v3 1/1 crypto/af_alg02: fixed read() being stuck
On kernels < 4.14 (missing commit 2d97591ef43d) reading from the request socket does not return and the testcase does not finish.
This fix moves the logic to a child thread in order for the parent to handle the timeout and report a message to the user.
---
Notes:
Hi Li,
> We could set LTP_ATTRIBUTE_UNUSED at the behind of unused parameter to get > rid of compiling warning
Thats very useful but I couldn't find it anywhere in the documentation. IMHO it should be put in there, because I stumbled upon this problem a couple of times.
Anyway, I implemented your suggestions. I hope it's an alright patch now. Thanks for your feedback!
Regards,
Christian
testcases/kernel/crypto/Makefile | 2 ++ testcases/kernel/crypto/af_alg02.c | 37 +++++++++++++++++++++++++++++++++++-- 2 files changed, 37 insertions(+), 2 deletions(-)
testcases/kernel/crypto/Makefile +2 -0
@@ -20,3 +20,5 @@ include $(top_srcdir)/include/mk/testcases.mk CFLAGS += -D_GNU_SOURCE include $(top_srcdir)/include/mk/generic_leaf_target.mk + +af_alg02: CFLAGS += -pthread
testcases/kernel/crypto/af_alg02.c +35 -2
@@ -7,23 +7,56 @@* Regression test for commit ecaaab564978 ("crypto: salsa20 - fix * blkcipher_walk API usage"), or CVE-2017-17805. This test verifies that an * empty message can be encrypted with Salsa20 without crashing the kernel. +* +* Fix for kernels < 4.14: +* With kernels missing commit 2d97591ef43d ("crypto: af_alg - consolidation +* of duplicate code") read() does not return in this situation. The call is +* now moved to a child thread in order to cancel it in case read() takes an +* unusual long amount of time. */ #include "tst_test.h" #include "tst_af_alg.h" +#include "tst_safe_pthread.h" +#include <pthread.h> +#include <errno.h> -static void run (void ) +void *verify_encrypt(void *arg LTP_ATTRIBUTE_UNUSED) {char buf[16 ];int reqfd =tst_alg_setup_reqfd ("skcipher" ,"salsa20" ,NULL ,16 ); +TST_CHECKPOINT_WAKE (0 ); +/* With the bug the kernel crashed here */ if (read (reqfd, buf,16 ) ==0 )tst_res (TPASS ,"Successfully \"encrypted\" an empty message" );else -tst_res (TBROK ,"read() didn't return 0" ); +tst_res (TFAIL ,"read() didn't return 0" ); +return arg; +} + +static void run (void ) +{ +pthread_t thr; + +pthread_setcancelstate (PTHREAD_CANCEL_ENABLE ,NULL ); +SAFE_PTHREAD_CREATE (&thr,NULL , verify_encrypt,NULL ); + +TST_CHECKPOINT_WAIT (0 ); + +while (pthread_kill (thr,0 ) !=ESRCH ) { +if (tst_timeout_remaining () <=10 ) { +pthread_cancel (thr); +tst_brk (TBROK , +"Timed out while reading from request socket." ); + } +usleep (1000 ); + } }static struct tst_test test = { .test_all = run, + .timeout =20 , + .needs_checkpoints =1 , };
Signature · 1 line
2.16.4
PATCH v3 1/1 crypto/af_alg02: fixed read() being stuck
Quoted 9 lines · Christian Amann
On kernels < 4.14 (missing commit
2d97591ef43d) reading from the request socket does not return and the testcase does not finish.This fix moves the logic to a child thread in order for the parent to handle the timeout and report a message to the user.
Signed-off-by Christian Amann <[email protected]>
---
Quoted 4 lines
Notes:
Hi Li,
Quoted 8 lines
> We could set
LTP_ATTRIBUTE_UNUSEDat the behind of unused parameter
to getQuoted 1 line
rid of compiling warning
Thats very useful but I couldn't find it anywhere in the documentation. IMHO it should be put in there, because I stumbled upon this problem a couple of times.
This is just a definition for variable attribute which supports by GNU compiler, I'm not sure if we should add it to LTP documents.
$ grep LTP_ATTRIBUTE_UNUSED . -r tst_common.h:25:#define LTP_ATTRIBUTE_UNUSED __attribute__((unused))
Quoted 66 lines
Anyway, I implemented your suggestions. I hope it's an alright patch
now.
Thanks for your feedback!Regards,
Christiantestcases/kernel/crypto/Makefile | 2 ++ testcases/kernel/crypto/af_alg02.c | 37 +++++++++++++++++++++++++++++++++++-- 2 files changed, 37 insertions(+), 2 deletions(-)testcases/kernel/crypto/Makefile +2
-0@@ -20,3 +20,5 @@ include $(top_srcdir)/include/mk/testcases.mk CFLAGS += -D_GNU_SOURCE include $(top_srcdir)/include/mk/generic_leaf_target.mk + +af_alg02: CFLAGS += -pthreadtestcases/kernel/crypto/af_alg02.c +0
-0@@ -7,23 +7,56 @@* Regression test for commit ecaaab564978 ("crypto: salsa20 - fix * blkcipher_walk API usage"), or CVE-2017-17805. This test verifies that an
* empty message can be encrypted with Salsa20 without crashing the
kernel.
+ *
+ * Fix for kernels < 4.14:
+ * With kernels missing commit2d97591ef43d("crypto:af_alg- consolidation
+ * of duplicate code")read()does not return in this situation. The call is
+ * now moved to a child thread in order to cancel it in caseread()takes an
+ * unusual long amount of time.
*/#include "
tst_test.h"
#include "tst_af_alg.h"testcases/kernel/crypto/af_alg02.c +6
-1+#include "tst_safe_pthread.h" +#include <pthread.h> +#include <errno.h> -static void run (void ) +void *verify_encrypt(void *arg LTP_ATTRIBUTE_UNUSED) {char buf[16 ];int reqfd =tst_alg_setup_reqfd ("skcipher" ,"salsa20" ,NULL ,16 ); +TST_CHECKPOINT_WAKE (0 ); +/* With the bug the kernel crashed here */ if (read (reqfd, buf,16 ) ==0 ) tst_res(TPASS, "Successfully \"encrypted\" an emptymessage");
else
testcases/kernel/crypto/af_alg02.c line 36
Quoted 3 lines
testcases/kernel/crypto/af_alg02.c +2
-1-tst_res (TBROK ,"read() didn't return 0" ); +tst_res (TFAIL ,"read() didn't return 0" ); +return arg;
Since arg has been marked as unused, so here better to return NULL.
Anyway, patch V3 looks good to me.
Signature · 5 lines
Regards, Li Wang -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.linux.it/pipermail/ltp/attachments/20190509/b74f1cc3/attachment.html>
PATCH v3 1/1 crypto/af_alg02: fixed read() being stuck
In reply to Christian AmannQuoted 5 lines
testcases/kernel/crypto/af_alg02.c +1
-1-static void run (void ) +void *verify_encrypt(void *arg LTP_ATTRIBUTE_UNUSED) {char buf[16 ];int reqfd =tst_alg_setup_reqfd ("skcipher" ,"salsa20" ,NULL ,16 );
testcases/kernel/crypto/af_alg02.c line 36
Quoted 10 lines
testcases/kernel/crypto/af_alg02.c +4
-1+TST_CHECKPOINT_WAKE (0 ); +/* With the bug the kernel crashed here */ if (read (reqfd, buf,16 ) ==0 )tst_res (TPASS ,"Successfully \"encrypted\" an empty message" );else -tst_res (TBROK ,"read() didn't return 0" ); +tst_res (TFAIL ,"read() didn't return 0" ); +return arg;
Actually there is no point in adding the LTP_ATTRIBUTE_UNUSED since you do return arg; at the end of the function.
So I've removed the LTP_ATTRIBUTE_UNUSED, changed the function to be static and pushed. thanks.
Quoted 32 lines
testcases/kernel/crypto/af_alg02.c +21
-0+} + +static void run (void ) +{ +pthread_t thr; + +pthread_setcancelstate (PTHREAD_CANCEL_ENABLE ,NULL ); +SAFE_PTHREAD_CREATE (&thr,NULL , verify_encrypt,NULL ); + +TST_CHECKPOINT_WAIT (0 ); + +while (pthread_kill (thr,0 ) !=ESRCH ) { +if (tst_timeout_remaining () <=10 ) { +pthread_cancel (thr); +tst_brk (TBROK , +"Timed out while reading from request socket." ); + } +usleep (1000 ); + } }static struct tst_test test = { .test_all = run, + .timeout =20 , + .needs_checkpoints =1 , };Signature · 5 lines
2.16.4 -- Mailing list info: https://lists.linux.it/listinfo/ltp
Signature · 2 lines
Cyril Hrubis [email protected]