cve-2017-17052: tolerate ENOMEM during test
PATCH 1/1 cve-2017-17052: tolerate ENOMEM during test
As each iteration of mmap_thread() grabs a fresh 16 MiB MAP_POPULATE region and never releases it. As the loop runs, those regions accumulate consuming both virtual address space and committed physical memory right away instead of lazily.
Easily mmap() fails with ENOMEM on smaller/limit RAM resource system.
Error Log:
cve-2017-17052.c:48: TBROK: mmap((nil),16777216,PROT_READ(1),32802,-1,0) failed: ENOMEM (12) tst_test.c:479: TINFO: Child process reported TBROK killing the test tst_test.c:1909: TINFO: Killed the leftover descendant processes
Consider there is no practical upper bound on this allocation pattern, so setting .test.min_mem_avail may not helps. Here we just tolerate ENOMEM during the mmap_thread() looping.
--- testcases/cve/cve-2017-17052.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
testcases/cve/cve-2017-17052.c +8 -1
@@ -44,9 +44,16 @@ static void cleanup(void)static void *mmap_thread (void *arg) { +void *ptr; +for (;;) { -SAFE_MMAP (NULL ,0x1000000 ,PROT_READ , + ptr =mmap (NULL ,0x1000000 ,PROT_READ ,MAP_POPULATE |MAP_ANONYMOUS |MAP_PRIVATE ,-1 ,0 ); + +if ((ptr ==MAP_FAILED ) && (errno ==ENOMEM )) { +usleep (1000 ); +continue ; + } }return arg;
Signature · 5 lines
2.52.0 -- Mailing list info: https://lists.linux.it/listinfo/ltp
Quoted 32 lines · Li Wang via ltp
As each iteration of
mmap_thread()grabs a fresh 16 MiBMAP_POPULATEregion and never releases it. As the loop runs, those regions accumulate consuming both virtual address space and committed physical memory right away instead of lazily.Easily
mmap()fails with ENOMEM on smaller/limit RAM resource system.Error Log:
cve-2017-17052.c:48: TBROK: mmap((nil),16777216,PROT_READ(1),32802,-1,0) failed: ENOMEM (12) tst_test.c:479: TINFO: Child process reported TBROK killing the test tst_test.c:1909: TINFO: Killed the leftover descendant processes
Consider there is no practical upper bound on this allocation pattern, so setting .test.min_mem_avail may not helps. Here we just tolerate ENOMEM during the
mmap_thread()looping.Signed-off-by Li Wang <[email protected]>--- testcases/cve/cve-2017-17052.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)testcases/cve/cve-2017-17052.c +1
-0@@ -44,9 +44,16 @@ static void cleanup(void)static void *mmap_thread (void *arg) { +void *ptr;
testcases/cve/cve-2017-17052.c static void cleanup(void) · line 56
Quoted 10 lines
testcases/cve/cve-2017-17052.c +7
-1+for (;;) { -SAFE_MMAP (NULL ,0x1000000 ,PROT_READ , + ptr =mmap (NULL ,0x1000000 ,PROT_READ ,MAP_POPULATE |MAP_ANONYMOUS |MAP_PRIVATE ,-1 ,0 ); + +if ((ptr ==MAP_FAILED ) && (errno ==ENOMEM )) { +usleep (1000 ); +continue ; + }
Oops, I forgot to handle the other failures, so the patch should be:
testcases/cve/cve-2017-17052.c +11 -1
@@ -44,9 +44,19 @@ static void cleanup(void)static void *mmap_thread (void *arg) { +void *ptr; +for (;;) { -SAFE_MMAP (NULL ,0x1000000 ,PROT_READ , + ptr =mmap (NULL ,0x1000000 ,PROT_READ ,MAP_POPULATE |MAP_ANONYMOUS |MAP_PRIVATE ,-1 ,0 ); + +if (ptr ==MAP_FAILED ) { +if (errno ==ENOMEM ) { +usleep (1000 ); +continue ; + } +tst_brk (TBROK |TTERRNO ,"Unexpected mmap() error" ); + } }return arg;
Signature · 6 lines
Regards, Li Wang -- Mailing list info: https://lists.linux.it/listinfo/ltp
Petr Vorel
Reviewed-byHi Li,
[ Cc others as it's a pre-release ]
Quoted 59 lines · Li Wang via ltp
Quoted 32 lines
As each iteration of
mmap_thread()grabs a fresh 16 MiBMAP_POPULATEregion and never releases it. As the loop runs, those regions accumulate consuming both virtual address space and committed physical memory right away instead of lazily.Easily
mmap()fails with ENOMEM on smaller/limit RAM resource system.Error Log:
cve-2017-17052.c:48: TBROK: mmap((nil),16777216,PROT_READ(1),32802,-1,0) failed: ENOMEM (12) tst_test.c:479: TINFO: Child process reported TBROK killing the test tst_test.c:1909: TINFO: Killed the leftover descendant processes
Consider there is no practical upper bound on this allocation pattern, so setting .test.min_mem_avail may not helps. Here we just tolerate ENOMEM during the
mmap_thread()looping.Signed-off-by Li Wang <[email protected]>--- testcases/cve/cve-2017-17052.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)testcases/cve/cve-2017-17052.c +1
-0@@ -44,9 +44,16 @@ static void cleanup(void)static void *mmap_thread (void *arg) { +void *ptr;testcases/cve/cve-2017-17052.c static void cleanup(void) · line 56
Quoted 10 lines
testcases/cve/cve-2017-17052.c +7
-1+for (;;) { -SAFE_MMAP (NULL ,0x1000000 ,PROT_READ , + ptr =mmap (NULL ,0x1000000 ,PROT_READ ,MAP_POPULATE |MAP_ANONYMOUS |MAP_PRIVATE ,-1 ,0 ); + +if ((ptr ==MAP_FAILED ) && (errno ==ENOMEM )) { +usleep (1000 ); +continue ; + }Oops, I forgot to handle the other failures, so the patch should be:
testcases/cve/cve-2017-17052.c +4
-1@@ -44,9 +44,19 @@ static void cleanup(void)static void *mmap_thread (void *arg) { +void *ptr; +for (;;) { -SAFE_MMAP (NULL ,0x1000000 ,PROT_READ , + ptr =mmap (NULL ,0x1000000 ,PROT_READ ,MAP_POPULATE |MAP_ANONYMOUS |MAP_PRIVATE ,-1 ,0 ); +
testcases/cve/cve-2017-17052.c static void cleanup(void) · line 62
Quoted 10 lines
testcases/cve/cve-2017-17052.c +7
-0+if (ptr ==MAP_FAILED ) { +if (errno ==ENOMEM ) { +usleep (1000 ); +continue ; + } +tst_brk (TBROK |TTERRNO ,"Unexpected mmap() error" ); + } }return arg;
Patched version LGTM.
Out of curiosity how much RAM is not enough for this? I tried 400 MB + eating other RAM but was not able to trigger.
Kind regards,
Petr
Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
Quoted 2 lines
Out of curiosity how much RAM is not enough for this? I tried 400 MB + eating other RAM but was not able to trigger.
It occurs (but rarely) in our CI jobs, and is hard to reproduce manually.
But if you limit resources in a cgroup or prolong the looping time for a while, that would be possible reproduce the fail.
Signature · 6 lines
Regards, Li Wang -- Mailing list info: https://lists.linux.it/listinfo/ltp
Cyril Hrubis
Reviewed-byHi!
testcases/cve/cve-2017-17052.c static void cleanup(void) · line 58
Quoted 11 lines
for (;;) {testcases/cve/cve-2017-17052.c +8
-1-SAFE_MMAP (NULL ,0x1000000 ,PROT_READ , + ptr =mmap (NULL ,0x1000000 ,PROT_READ ,MAP_POPULATE |MAP_ANONYMOUS |MAP_PRIVATE ,-1 ,0 ); + +if (ptr ==MAP_FAILED ) { +if (errno ==ENOMEM ) { +usleep (1000 ); +continue ; + } +tst_brk (TBROK |TTERRNO ,"Unexpected mmap() error" );
Maybe this part would be more readable with an else branch:
if (errno == ENOMEM) usleep(1000); else tst_brk(TBROK | TTERRNO, "mmap() failed");
Otherwise:
Quoted 14 lines
+ }
} return arg;Signature · 6 lines
Regards, Li Wang -- Mailing list info: https://lists.linux.it/listinfo/ltp
Signature · 5 lines
Cyril Hrubis [email protected] -- Mailing list info: https://lists.linux.it/listinfo/ltp
Quoted 13 lines · Cyril Hrubis
Hi!
testcases/cve/cve-2017-17052.c static void cleanup(void)
Quoted 4 lines
for (;;) {testcases/cve/cve-2017-17052.c +1
-1- SAFE_MMAP(NULL, 0x1000000, PROT_READ, + ptr = mmap(NULL, 0x1000000, PROT_READ, MAP_POPULATE|MAP_ANONYMOUS|MAP_PRIVATE,-1, 0);
testcases/cve/cve-2017-17052.c static void cleanup(void)
Quoted 7 lines
testcases/cve/cve-2017-17052.c +7
-0+ +if (ptr ==MAP_FAILED ) { +if (errno ==ENOMEM ) { +usleep (1000 ); +continue ; + } + tst_brk(TBROK | TTERRNO, "Unexpected mmap()
Quoted 9 lines
error");
Maybe this part would be more readable with an else branch:
if (errno == ENOMEM) usleep(1000); else tst_brk(TBROK | TTERRNO, "mmap() failed");
Good suggestion, patch merged like this. Thanks!
Signature · 5 lines
Regards, Li Wang -- Mailing list info: https://lists.linux.it/listinfo/ltp