Message ID | 2e51e5453562001b5b6992ccc897d9177e6673ca.1571053661.git.jstancek@redhat.com |
---|---|
State | Deferred, archived |
Headers | show |
Series | mlock_8-1, munlock_10-1: don't use LONG_MAX as invalid pointer | expand |
Hi! > arm64 kernel with commit 057d3389108e ("mm: untag user pointers passed to > memory syscalls") will assume this is tagged user pointer and will > attempt to clear first byte. This leads to test hitting different error > (EINVAL instead of ENOMEM). > > Add a helper function which provides invalid/unused pointer from user > address space. > > Signed-off-by: Jan Stancek <jstancek@redhat.com> > --- > .../conformance/interfaces/mlock/8-1.c | 11 ++-------- > .../conformance/interfaces/munlock/10-1.c | 11 ++-------- > .../open_posix_testsuite/include/invalid_helpers.h | 25 ++++++++++++++++++++++ > 3 files changed, 29 insertions(+), 18 deletions(-) > create mode 100644 testcases/open_posix_testsuite/include/invalid_helpers.h > > diff --git a/testcases/open_posix_testsuite/conformance/interfaces/mlock/8-1.c b/testcases/open_posix_testsuite/conformance/interfaces/mlock/8-1.c > index d9569008b82b..5883ff8ec162 100644 > --- a/testcases/open_posix_testsuite/conformance/interfaces/mlock/8-1.c > +++ b/testcases/open_posix_testsuite/conformance/interfaces/mlock/8-1.c > @@ -20,22 +20,15 @@ > #include <errno.h> > #include <limits.h> > #include "posixtest.h" > +#include "invalid_helpers.h" > > #define BUFSIZE 8 > > int main(void) > { > int result; > - long page_size; > - void *page_ptr; > + void *page_ptr = get_invalid_ptr(); > > - page_size = sysconf(_SC_PAGESIZE); > - if (errno) { > - perror("An error occurs when calling sysconf()"); > - return PTS_UNRESOLVED; > - } > - > - page_ptr = (void *)(LONG_MAX - (LONG_MAX % page_size)); > result = mlock(page_ptr, BUFSIZE); > > if (result == -1 && errno == ENOMEM) { > diff --git a/testcases/open_posix_testsuite/conformance/interfaces/munlock/10-1.c b/testcases/open_posix_testsuite/conformance/interfaces/munlock/10-1.c > index cf870289b512..1bc3d8015808 100644 > --- a/testcases/open_posix_testsuite/conformance/interfaces/munlock/10-1.c > +++ b/testcases/open_posix_testsuite/conformance/interfaces/munlock/10-1.c > @@ -20,22 +20,15 @@ > #include <errno.h> > #include <limits.h> > #include "posixtest.h" > +#include "invalid_helpers.h" > > #define BUFSIZE 8 > > int main(void) > { > int result; > - long page_size; > - void *page_ptr; > + void *page_ptr = get_invalid_ptr(); > > - page_size = sysconf(_SC_PAGESIZE); > - if (errno) { > - perror("An error occurs when calling sysconf()"); > - return PTS_UNRESOLVED; > - } > - > - page_ptr = (void *)(LONG_MAX - (LONG_MAX % page_size)); > result = munlock(page_ptr, BUFSIZE); > > if (result == -1 && errno == ENOMEM) { > diff --git a/testcases/open_posix_testsuite/include/invalid_helpers.h b/testcases/open_posix_testsuite/include/invalid_helpers.h > new file mode 100644 > index 000000000000..7882c783abf6 > --- /dev/null > +++ b/testcases/open_posix_testsuite/include/invalid_helpers.h > @@ -0,0 +1,25 @@ > +/* SPDX-License-Identifier: GPL-2.0-or-later > + * Copyright (c) 2019 Linux Test Project > + */ > + > +#ifndef __INVALID_HELPERS_H__ > +#define __INVALID_HELPERS_H__ ^ Strictly speaking the double underscores are reserved for low level code i.e. libc and related code so we should avoid them. > +#include <stdlib.h> > +#include "safe_helpers.h" > + > +#define DISTANT_AREA_SIZE (128*1024*1024) > + > +static void *get_invalid_ptr(void) > +{ > + int page_size; > + void *ptr; > + > + page_size = sysconf(_SC_PAGESIZE); > + SAFE_PFUNC(posix_memalign(&ptr, page_size, DISTANT_AREA_SIZE)); > + free(ptr); > + > + return (char *)ptr + (DISTANT_AREA_SIZE / 2); > +} Maybe it's worth of a comment what is the idea behind this code here. > +#endif > -- > 1.8.3.1 > > > -- > Mailing list info: https://lists.linux.it/listinfo/ltp
Hi Jan, > arm64 kernel with commit 057d3389108e ("mm: untag user pointers passed to > memory syscalls") will assume this is tagged user pointer and will > attempt to clear first byte. This leads to test hitting different error > (EINVAL instead of ENOMEM). > Add a helper function which provides invalid/unused pointer from user > address space. > Signed-off-by: Jan Stancek <jstancek@redhat.com> Reviewed-by: Petr Vorel <pvorel@suse.cz> Agree with Cyril points (double underscores at the beginning are reserved for libc and putting very brief info at get_invalid_ptr((). Kind regards, Petr
----- Original Message ----- > Hi Jan, > > > arm64 kernel with commit 057d3389108e ("mm: untag user pointers passed to > > memory syscalls") will assume this is tagged user pointer and will > > attempt to clear first byte. This leads to test hitting different error > > (EINVAL instead of ENOMEM). > > > Add a helper function which provides invalid/unused pointer from user > > address space. > > > Signed-off-by: Jan Stancek <jstancek@redhat.com> > Reviewed-by: Petr Vorel <pvorel@suse.cz> Thanks, discussion in: http://lists.linux.it/pipermail/ltp/2019-October/013993.html is still going, so maybe this will be fixed on kernel side after all.
Hi Jan, > > > arm64 kernel with commit 057d3389108e ("mm: untag user pointers passed to > > > memory syscalls") will assume this is tagged user pointer and will > > > attempt to clear first byte. This leads to test hitting different error > > > (EINVAL instead of ENOMEM). > > > Add a helper function which provides invalid/unused pointer from user > > > address space. > > > Signed-off-by: Jan Stancek <jstancek@redhat.com> > > Reviewed-by: Petr Vorel <pvorel@suse.cz> > Thanks, discussion in: > http://lists.linux.it/pipermail/ltp/2019-October/013993.html > is still going, so maybe this will be fixed on kernel side > after all. Thanks for info, I didn't notice this is related to your patch. Kind regards, Petr
Hi Jan, I propose to reorg this patch with review comments (maybe apply it after the new release). Any thought?
----- Original Message ----- > Hi Jan, > > I propose to reorg this patch with review comments (maybe apply it after > the new release). > > Any thought? This has been fixed in kernel by: 597399d0cb91 ("arm64: tags: Preserve tags for addresses translated via TTBR1") d0022c0ef29b ("arm64: memory: Add missing brackets to untagged_addr() macro") I'd keep the test as is, maybe add comment with above commits as a hint for future.
Jan Stancek <jstancek@redhat.com> wrote: ... > > I propose to reorg this patch with review comments (maybe apply it after > > the new release). > > > > Any thought? > > This has been fixed in kernel by: > 597399d0cb91 ("arm64: tags: Preserve tags for addresses translated via > TTBR1") > d0022c0ef29b ("arm64: memory: Add missing brackets to untagged_addr() > macro") > Thanks for the info, then we could filter this FAIL by kernel version. > > I'd keep the test as is, maybe add comment with above commits as a hint > for future. > +1
diff --git a/testcases/open_posix_testsuite/conformance/interfaces/mlock/8-1.c b/testcases/open_posix_testsuite/conformance/interfaces/mlock/8-1.c index d9569008b82b..5883ff8ec162 100644 --- a/testcases/open_posix_testsuite/conformance/interfaces/mlock/8-1.c +++ b/testcases/open_posix_testsuite/conformance/interfaces/mlock/8-1.c @@ -20,22 +20,15 @@ #include <errno.h> #include <limits.h> #include "posixtest.h" +#include "invalid_helpers.h" #define BUFSIZE 8 int main(void) { int result; - long page_size; - void *page_ptr; + void *page_ptr = get_invalid_ptr(); - page_size = sysconf(_SC_PAGESIZE); - if (errno) { - perror("An error occurs when calling sysconf()"); - return PTS_UNRESOLVED; - } - - page_ptr = (void *)(LONG_MAX - (LONG_MAX % page_size)); result = mlock(page_ptr, BUFSIZE); if (result == -1 && errno == ENOMEM) { diff --git a/testcases/open_posix_testsuite/conformance/interfaces/munlock/10-1.c b/testcases/open_posix_testsuite/conformance/interfaces/munlock/10-1.c index cf870289b512..1bc3d8015808 100644 --- a/testcases/open_posix_testsuite/conformance/interfaces/munlock/10-1.c +++ b/testcases/open_posix_testsuite/conformance/interfaces/munlock/10-1.c @@ -20,22 +20,15 @@ #include <errno.h> #include <limits.h> #include "posixtest.h" +#include "invalid_helpers.h" #define BUFSIZE 8 int main(void) { int result; - long page_size; - void *page_ptr; + void *page_ptr = get_invalid_ptr(); - page_size = sysconf(_SC_PAGESIZE); - if (errno) { - perror("An error occurs when calling sysconf()"); - return PTS_UNRESOLVED; - } - - page_ptr = (void *)(LONG_MAX - (LONG_MAX % page_size)); result = munlock(page_ptr, BUFSIZE); if (result == -1 && errno == ENOMEM) { diff --git a/testcases/open_posix_testsuite/include/invalid_helpers.h b/testcases/open_posix_testsuite/include/invalid_helpers.h new file mode 100644 index 000000000000..7882c783abf6 --- /dev/null +++ b/testcases/open_posix_testsuite/include/invalid_helpers.h @@ -0,0 +1,25 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later + * Copyright (c) 2019 Linux Test Project + */ + +#ifndef __INVALID_HELPERS_H__ +#define __INVALID_HELPERS_H__ + +#include <stdlib.h> +#include "safe_helpers.h" + +#define DISTANT_AREA_SIZE (128*1024*1024) + +static void *get_invalid_ptr(void) +{ + int page_size; + void *ptr; + + page_size = sysconf(_SC_PAGESIZE); + SAFE_PFUNC(posix_memalign(&ptr, page_size, DISTANT_AREA_SIZE)); + free(ptr); + + return (char *)ptr + (DISTANT_AREA_SIZE / 2); +} + +#endif
arm64 kernel with commit 057d3389108e ("mm: untag user pointers passed to memory syscalls") will assume this is tagged user pointer and will attempt to clear first byte. This leads to test hitting different error (EINVAL instead of ENOMEM). Add a helper function which provides invalid/unused pointer from user address space. Signed-off-by: Jan Stancek <jstancek@redhat.com> --- .../conformance/interfaces/mlock/8-1.c | 11 ++-------- .../conformance/interfaces/munlock/10-1.c | 11 ++-------- .../open_posix_testsuite/include/invalid_helpers.h | 25 ++++++++++++++++++++++ 3 files changed, 29 insertions(+), 18 deletions(-) create mode 100644 testcases/open_posix_testsuite/include/invalid_helpers.h